{ "name": "Data Sorcerers", "origin": "https://data-sorcerers.com", "pages": [ { "title": "Data Sorcerers", "url": "https://data-sorcerers.com/", "summary": "Data Sorcerers builds AI applications, data intelligence platforms, and software that help teams turn complex information into clear decisions.", "sections": [] }, { "title": "Community", "url": "https://data-sorcerers.com/community/", "summary": "Learn together, exchange ideas, and put AI into practice. Discover the communities and initiatives in the Data Sorcerers ecosystem.", "sections": [ { "heading": "Data Sorcerers Community", "body": "A collaborative AI and data community in Indonesia. Learn with others, experiment with new ideas, contribute to research, and turn what you know into working projects.", "linkSlug": "data-sorcerers-community" }, { "heading": "HerAI Fellowship", "body": "An initiative by Data Sorcerers and Womanhub.id helping Indonesian women build skills in AI, data science, and engineering through structured learning, mentorship, and practical projects.", "linkSlug": "herai" }, { "heading": "Sorcery AI Racing League", "body": "A virtual autonomous racing competition where you build the intelligence behind the wheel. Explore classical control or train a neural driver, test your approach, and put your code on the track.", "linkSlug": "sarl-sorcery-ai-racing-league" } ] }, { "title": "Products", "url": "https://data-sorcerers.com/products/", "summary": "Find intelligence platforms, applied AI, practical learning, certification, and language model integration in one ecosystem.", "sections": [ { "heading": "Data Intelligence Platform", "linkSlug": "data-intelligence-platform", "body": "Decision-grade intelligence modules for monitoring sectors, risk, reputation, markets, and operational signals.", "items": [ "Public Sector & Crisis Intelligence", "Financial & Investment Intelligence", "Executive & Competitive Intelligence", "Geopolitical Risk Intelligence", "Reputation Risk Intelligence", "Political Risk Monitoring", "Supply Chain & Geospatial Risk", "Cyber & Digital Risk", "Enterprise Risk Early Warning", "Healthcare & Public Health Intelligence", "Retail & Consumer Brand Intelligence", "Misinformation & Disinformation Monitoring", "Deepfake & Synthetic Media Risk", "Investment & Portfolio Intelligence" ] }, { "heading": "Applied AI Products", "linkSlug": "applied-ai-products", "body": "Applied AI product directions that can be configured for domain workflows, enterprise operations, public service, and specialist teams.", "items": [ "AI for Finance", "AI for Healthcare", "AI for Social", "AI for Business", "AI for Education", "AI for Government", "AI for Climate & Environment", "AI for Cybersecurity", "AI for Law & Compliance", "AI for Research & Science", "AI for Human Resources", "AI for Media & Creative", "AI for Transportation & Logistics", "AI for Industry & Manufacturing", "AI for Agriculture", "AI for Personal Productivity" ] }, { "heading": "Pyronyx Academy", "body": "Explore practical learning in AI, data, and software. Visit Pyronyx Academy for its learning programs and training options.", "linkSlug": "pyronyx-academy" }, { "heading": "Pyronyx Certification", "body": "Explore the certification pathway that fits your skills and goals. Visit Pyronyx Certification for program and assessment details.", "linkSlug": "pyronyx-certification" }, { "heading": "LLM Integration", "body": "Connect language models to your applications and workflows. Visit Arutala to explore LLM integration.", "linkSlug": "llm-integration" } ] }, { "title": "Data Intelligence Platform", "url": "https://data-sorcerers.com/data-intelligence-platform/", "summary": "A live intelligence layer that turns fragmented public, market, media, geospatial, and internal signals into decision-ready monitoring, dashboards, and alerts.", "sections": [ { "heading": "Who it is for", "body": "Leadership teams, analysts, public-sector groups, risk teams, investment teams, and operators who need to monitor fast-changing signals before they become expensive surprises.", "items": [ "Executive and strategy teams", "Risk, policy, and crisis units", "Market and competitive intelligence teams", "Operations and supply-chain leaders" ] }, { "heading": "What we build", "body": "We design the data pipelines, intelligence taxonomy, scoring logic, dashboards, alerts, and reporting workflows needed to move from raw signal to confident action.", "items": [ "Intelligence dashboards and command centers", "Early-warning alert systems", "Risk scoring and signal extraction", "Knowledge graphs and briefing workflows", "Automated executive intelligence reports" ] }, { "heading": "Related modules", "body": "The platform can start narrow and expand into sector-specific modules as the organization matures.", "items": [ "Public Sector & Crisis Intelligence", "Financial & Investment Intelligence", "Executive & Competitive Intelligence", "Geopolitical Risk Intelligence", "Reputation Risk Intelligence", "Political Risk Monitoring", "Supply Chain & Geospatial Risk", "Cyber & Digital Risk", "Enterprise Risk Early Warning", "Healthcare & Public Health Intelligence", "Retail & Consumer Brand Intelligence", "Misinformation & Disinformation Monitoring", "Deepfake & Synthetic Media Risk", "Investment & Portfolio Intelligence" ] }, { "heading": "Typical outputs", "body": "A working engagement usually ships a monitoring dashboard, priority alerts, data connectors, decision workflow, and repeatable intelligence briefing format." } ] }, { "title": "Applied AI Products", "url": "https://data-sorcerers.com/applied-ai-products/", "summary": "Domain-specific AI products for research, analysis, document intelligence, workflow automation, and decision support.", "sections": [ { "heading": "Who it is for", "body": "Teams that want AI embedded into real work—not a generic chatbot detached from data, process, governance, and user needs.", "items": [ "Product and innovation teams", "Analysts and research teams", "Customer operations and back-office teams", "Legal, compliance, HR, finance, healthcare, and public-sector teams" ] }, { "heading": "What we build", "body": "We build AI assistants, RAG systems, workflow agents, model-powered dashboards, evaluation pipelines, and human-in-the-loop systems that support real operational decisions.", "items": [ "RAG and enterprise knowledge-base apps", "AI research and reporting assistants", "Document intelligence workflows", "Monitoring and automation agents", "Evaluation, guardrails, and AI safety layers" ] }, { "heading": "Related AI products", "body": "Applied AI products can be shaped around industry needs, department workflows, or a specific high-value use case.", "items": [ "AI for Finance", "AI for Healthcare", "AI for Social", "AI for Business", "AI for Education", "AI for Government", "AI for Climate & Environment", "AI for Cybersecurity", "AI for Law & Compliance", "AI for Research & Science", "AI for Human Resources", "AI for Media & Creative", "AI for Transportation & Logistics", "AI for Industry & Manufacturing", "AI for Agriculture", "AI for Personal Productivity" ] }, { "heading": "Typical outputs", "body": "A practical delivery creates a tested prototype, evaluation dataset, safety rules, working interface, deployment plan, and roadmap for production hardening." } ] }, { "title": "Different strengths. A shared purpose.", "url": "https://data-sorcerers.com/partners/", "summary": "We welcome universities, businesses, public institutions, media, legal professionals, and technology teams who want to build a clearly scoped collaboration around data and AI.", "sections": [ { "id": "approach", "heading": "Start with what we can do together", "body": "A useful partnership connects complementary strengths with a specific outcome. Begin with the people the work should help, the contribution each team can make, and the responsibilities needed to deliver it. The categories below describe ways to start that conversation." }, { "id": "university-partners", "heading": "University Partners", "body": "We welcome conversations with universities and academic departments about practical learning, research exchange, and programs that connect students with data and AI.", "linkSlug": "university-partners" }, { "id": "company-partners", "heading": "Company Partners", "body": "Explore joint delivery, learning programs, and product collaboration with a clearly defined business purpose.", "linkSlug": "company-partners" }, { "id": "legal-partners", "heading": "Legal Partners", "body": "We welcome discussions with qualified legal professionals about the questions that arise when organizations build and use data, AI, and digital products.", "linkSlug": "legal-partners" }, { "id": "government-partners", "heading": "Government Partners", "body": "We welcome dialogue with public institutions about data capability, research, and digital tools grounded in the needs of the people they serve.", "linkSlug": "government-partners" }, { "id": "media-partners", "heading": "Media Partners", "body": "Explore editorial conversations, interviews, event coverage, and educational formats about data, AI, and the people working with them.", "linkSlug": "media-partners" }, { "id": "technology-alliance-partners", "heading": "Technology Alliance Partners", "body": "Explore technical integrations, complementary capabilities, and shared engineering work around a specific user need.", "linkSlug": "technology-alliance-partners" }, { "id": "next-step", "heading": "From an introduction to a shared plan", "body": "Send a short proposal describing your organization, the opportunity, intended participants or users, timeline, and contact person. We will discuss fit, clarify responsibilities, and agree on a scope before announcing or starting a collaboration.", "links": [ { "label": "Propose a partnership", "href": "mailto:contact@data-sorcerers.com?subject=Partnership%20proposal" } ] } ] }, { "title": "Bring research, learning, and practice closer together.", "url": "https://data-sorcerers.com/university-partners/", "summary": "We welcome conversations with universities and academic departments about practical learning, research exchange, and programs that connect students with data and AI.", "sections": [ { "id": "chapter-1", "heading": "A shared question is a useful starting point", "body": "A collaboration can begin with a lecturer’s research question, a student community’s learning needs, or a department exploring practical applications. We would first work together to understand the academic context, intended participants, and what a useful contribution would look like." }, { "id": "chapter-2", "heading": "Possible ways to collaborate", "body": "Depending on the institution’s priorities, a proposal could include guest sessions, project-based workshops, mentoring, research discussions, or a jointly scoped student activity. These are potential collaboration formats; the scope, supervision, and resources would be agreed for each program." }, { "id": "chapter-3", "heading": "Design the experience around participants", "body": "Define prerequisites, accessibility needs, learning goals, the format, and the responsibilities of both teams. For projects involving research or student work, agree on data access, attribution, publication, intellectual property, and any necessary institutional review before work begins." }, { "id": "chapter-4", "heading": "Start with a short academic brief", "body": "Share your institution and department, the intended audience, proposed topic, preferred dates, format, and a contact person. If you already have a program concept, include the learning or research outcome you want to pursue. We can then discuss fit and a realistic next step.", "links": [ { "label": "Discuss this partnership", "href": "mailto:contact@data-sorcerers.com?subject=University%20Partners%20inquiry" }, { "label": "Contact options", "href": "/contact-us/" } ] } ] }, { "title": "Build a collaboration around work worth doing.", "url": "https://data-sorcerers.com/company-partners/", "summary": "Explore joint delivery, learning programs, and product collaboration with a clearly defined business purpose.", "sections": [ { "id": "chapter-1", "heading": "Look for complementary strengths", "body": "A productive partnership should explain what each team contributes and why working together is useful. That might mean combining domain expertise with engineering, connecting an existing product to an intelligence workflow, or preparing a team to use new tools. The starting point is a shared problem, not a logo exchange." }, { "id": "chapter-2", "heading": "Possible collaboration paths", "body": "We are open to discussing jointly scoped solutions, product integrations, team learning, and business-development collaboration. A commercial proposal should identify the users, expected deliverables, available resources, and how success will be assessed. No specific delivery or referral arrangement is implied before agreement." }, { "id": "chapter-3", "heading": "Make responsibilities clear", "body": "Before delivery, align on decision owners, communication, customer relationships, security responsibilities, commercial terms, and support. Where customer data or confidential information is involved, establish the appropriate access and handling arrangements before sharing it." }, { "id": "chapter-4", "heading": "Tell us where the opportunity sits", "body": "Send a short description of your company, the customer or operational need, your proposed contribution, and the timeline. We will discuss whether there is a useful fit and what information is needed to develop a concrete scope.", "links": [ { "label": "Discuss this partnership", "href": "mailto:contact@data-sorcerers.com?subject=Company%20Partners%20inquiry" }, { "label": "Contact options", "href": "/contact-us/" } ] } ] }, { "title": "Connect technical decisions with informed legal review.", "url": "https://data-sorcerers.com/legal-partners/", "summary": "We welcome discussions with qualified legal professionals about the questions that arise when organizations build and use data, AI, and digital products.", "sections": [ { "id": "chapter-1", "heading": "Bring expertise into the right conversation", "body": "Technology projects can raise questions about contracts, data handling, ownership, licensing, and responsibilities between parties. A legal collaboration would connect appropriately qualified counsel with the technical context needed for that counsel to advise. This page describes a potential partnership path, not legal advice." }, { "id": "chapter-2", "heading": "Possible areas of collaboration", "body": "A scoped engagement could involve reviewing technology agreements, discussing intellectual-property arrangements, or developing educational sessions with legal specialists. The jurisdiction, professional responsibilities, and limits of any engagement must be established by the relevant parties." }, { "id": "chapter-3", "heading": "Keep professional roles explicit", "body": "Data Sorcerers can contribute technical context and implementation knowledge. Legal advice remains the responsibility of qualified professionals retained for that purpose. A proposed partnership should address conflicts, confidentiality, client relationships, and the handling of privileged or sensitive material where applicable." }, { "id": "chapter-4", "heading": "Begin without confidential documents", "body": "Introduce your practice, areas of expertise, relevant jurisdiction, and the collaboration you have in mind. An initial inquiry should not include privileged client material. The parties can agree on suitable arrangements before discussing confidential details.", "links": [ { "label": "Discuss this partnership", "href": "mailto:contact@data-sorcerers.com?subject=Legal%20Partners%20inquiry" }, { "label": "Contact options", "href": "/contact-us/" } ] } ] }, { "title": "Explore technology around a clear public-service need.", "url": "https://data-sorcerers.com/government-partners/", "summary": "We welcome dialogue with public institutions about data capability, research, and digital tools grounded in the needs of the people they serve.", "sections": [ { "id": "chapter-1", "heading": "Start with the service and its users", "body": "A discussion should begin with a public-service problem, the people affected, and the institution’s operating context. Data availability, existing workflows, access requirements, and the capacity to maintain a solution are part of that conversation. A demonstration alone cannot establish public-service readiness." }, { "id": "chapter-2", "heading": "Possible collaboration formats", "body": "Potential activities include capability-building sessions, research exchange, discovery work, or a carefully scoped pilot. Any collaboration would depend on institutional approval, applicable procurement processes, and an agreed scope. This page does not indicate an appointment, endorsement, or awarded government contract." }, { "id": "chapter-3", "heading": "Design for accountability", "body": "A proposal should explain human oversight, accessibility, information handling, evaluation, and the responsibilities of the operating team. Where technology informs consequential decisions, establish how people can review the output and how concerns will be raised within the institution’s processes." }, { "id": "chapter-4", "heading": "Share the institutional context", "body": "Contact us with your unit, the service challenge, intended users, project stage, relevant constraints, and a focal point. Please use appropriate institutional channels and avoid sending restricted data in an introductory message.", "links": [ { "label": "Discuss this partnership", "href": "mailto:contact@data-sorcerers.com?subject=Government%20Partners%20inquiry" }, { "label": "Contact options", "href": "/contact-us/" } ] } ] }, { "title": "Make technical stories understandable and well sourced.", "url": "https://data-sorcerers.com/media-partners/", "summary": "Explore editorial conversations, interviews, event coverage, and educational formats about data, AI, and the people working with them.", "sections": [ { "id": "chapter-1", "heading": "Context makes a technology story useful", "body": "Technical language can obscure the question a reader actually cares about: what changes, for whom, and with what limits? We welcome media conversations that connect technology with its practical context and distinguish a demonstrated result from an ambition or proposal." }, { "id": "chapter-2", "heading": "Possible editorial collaboration", "body": "A proposal might involve an expert conversation, an educational segment, a community-event story, or a request for factual background. Editorial coverage, sponsored work, and event partnerships should be discussed as distinct arrangements with clear expectations and appropriate disclosure." }, { "id": "chapter-3", "heading": "Protect the integrity of the story", "body": "Agree on the subject, source material, deadlines, attribution, and permissions for any supplied assets. We can help check technical context and company facts, while an editorial outlet retains responsibility for its own reporting and editorial decisions. Access to a conversation does not imply endorsement." }, { "id": "chapter-4", "heading": "Send an editorial brief", "body": "Include your publication or channel, intended audience, topic, format, deadline and timezone, and the support you are requesting. If you need an interview or image permissions, mention that in the first message so we can route the request.", "links": [ { "label": "Discuss this partnership", "href": "mailto:contact@data-sorcerers.com?subject=Media%20Partners%20inquiry" }, { "label": "Contact options", "href": "/contact-us/" } ] } ] }, { "title": "Make the connection useful—and operable.", "url": "https://data-sorcerers.com/technology-alliance-partners/", "summary": "Explore technical integrations, complementary capabilities, and shared engineering work around a specific user need.", "sections": [ { "id": "chapter-1", "heading": "Begin with the workflow between systems", "body": "A technology alliance should explain how two capabilities work together and what improves for the user. Start with the data flow, the interface, and the operational responsibility on each side. Product compatibility is something to establish through testing, not a claim to make from a diagram." }, { "id": "chapter-2", "heading": "Possible technical collaboration", "body": "We are open to discussing API integrations, AI and data workflows, deployment patterns, and jointly scoped technical validation. A useful proposal names the product or component, supported environments, relevant interfaces, and the problem the integration would solve." }, { "id": "chapter-3", "heading": "Plan beyond the first successful request", "body": "Agree on authentication, access scopes, version compatibility, performance expectations, incident handling, and maintenance ownership. Define how breaking changes are communicated and what happens when a dependency is unavailable. The integration needs an operating plan as well as a prototype." }, { "id": "chapter-4", "heading": "Bring a concise technical brief", "body": "Share the platform, intended workflow, documentation links, integration requirements, and a technical contact. Do not include API keys or credentials. If a sandbox is needed, we can discuss its scope and access arrangements after the initial review.", "links": [ { "label": "Discuss this partnership", "href": "mailto:contact@data-sorcerers.com?subject=Technology%20Alliance%20Partners%20inquiry" }, { "label": "Contact options", "href": "/contact-us/" } ] } ] }, { "title": "More people able to build with data and AI.", "url": "https://data-sorcerers.com/company/", "summary": "A student finding a first project. A business making sense of its information. A team turning an idea into a working product. These are the people and questions that connect our work.", "sections": [ { "id": "our-story", "heading": "A shared belief, put into practice", "body": "Useful technology begins with access: to knowledge, to people who can help, and to the opportunity to try. At Data Sorcerers, community learning and technical delivery belong in the same story. One helps people build capability; the other brings that capability to practical problems.", "linkSlug": "about-us" }, { "id": "impacts", "heading": "Learning that reaches beyond one room", "body": "Our programs have reached more than 10,000 school and university students across Indonesia and Southeast Asia. Gatherings, talks, mentorship, and fellowships offer different ways to participate. University collaborations connect us with more than 15 Indonesian cities.", "linkSlug": "impacts" }, { "id": "customers", "heading": "Working alongside businesses", "body": "Our support reaches 15 Indonesian MSMEs, three medium enterprises in Singapore, and five Indonesian private companies. Their contexts differ. Our starting point is consistent: understand the decision, the people involved, and what a useful solution needs to do.", "linkSlug": "customers" }, { "id": "events", "heading": "Places to meet, learn, and return", "body": "Sorcery Gathering has grown through editions 1–11, alongside Sorcery Talks Series, Sorcery AI Mentorship 2025, and HerAI Fellowship 2026. These programs connect conversations with longer opportunities to learn.", "linkSlug": "events" }, { "id": "leadership", "heading": "Leadership is a responsibility to the work", "body": "Setting direction means making trade-offs clear and keeping technical work connected to the people it serves. Learning, research, engineering, and community all need a voice in those decisions.", "linkSlug": "leadership" }, { "id": "news", "heading": "Milestones with their context", "body": "Follow company updates and public recognition, including the 2025 PKB Rising Changemaker Award nomination. Each story explains what happened, when it happened, and where the information comes from.", "linkSlug": "news" }, { "id": "careers", "heading": "The next chapter needs different strengths", "body": "Explore 18 open remote internships across research, engineering, design, community, education, and business. All roles coordinate in GMT+7. Find responsibilities and application details on our Careers portal.", "linkSlug": "careers" }, { "id": "reviews", "heading": "Listening is part of the work", "body": "Participant and customer feedback helps us understand what was useful, what was unclear, and what should change. Share your experience or talk to us about references relevant to your project.", "linkSlug": "reviews" }, { "id": "contact-us", "heading": "Bring us the question you are working on", "body": "Whether you want to build a product, host a learning program, collaborate with a university, or support a community initiative, start with a conversation.", "linkSlug": "contact-us" } ] }, { "title": "Opening doors to data and AI.", "url": "https://data-sorcerers.com/impacts/", "summary": "More than 10,000 school and university students reached across Indonesia and Southeast Asia. Our impact connects access to learning, sustained mentoring, and practical support for businesses.", "sections": [ { "id": "access", "heading": "An entry point is only the beginning", "body": "An online session can be a first introduction to AI. A gathering makes room for questions and new connections. Mentorship creates time to work through the next challenge. Our programs offer these different entry points so that participation can continue beyond a single event.", "items": [ "Sorcery Gathering editions 1–11: a recurring community meeting point.", "Sorcery Talks Series: conversations around data, AI, and the work surrounding them.", "Sorcery AI Mentorship 2025: certification and mentoring grants valued at more than IDR 2.2 billion.", "HerAI Fellowship 2026: a dedicated fellowship within the HerAI community." ], "links": [ { "label": "Meet the programs", "href": "/events/" }, { "label": "Join our communities", "href": "/community/" } ] }, { "id": "campuses", "heading": "Across disciplines. Across cities.", "body": "Data and AI questions appear in many fields, not just computer science. We collaborate with different departments at universities across more than 15 Indonesian cities. Online and offline formats allow these conversations to reach students in different learning environments, including across Southeast Asia.", "links": [ { "label": "Bring a program to your campus", "href": "mailto:contact@data-sorcerers.com?subject=University%20program%20collaboration" } ] }, { "id": "incubation", "heading": "Giving early ideas room to develop", "body": "In 2025, Data Sorcerers supported five MSMEs and startups through its incubation work. In 2026, that support includes ten startups. These cohorts reflect a continuing commitment to work alongside early businesses as they develop their ideas.", "items": [ "2025 · 5 MSMEs and startups supported through incubation.", "2026 · 10 startups supported, as of 31 October." ], "links": [ { "label": "Discuss startup collaboration", "href": "/contact-us/" } ] }, { "id": "business-support", "heading": "Taking knowledge into everyday business", "body": "Alongside education and incubation, we have helped 15 MSMEs and five private companies in Indonesia, and three medium enterprises in Singapore. These are separate support categories; businesses in incubation may also appear in the broader business-support figures.", "links": [ { "label": "How we work with customers", "href": "/customers/" } ] }, { "id": "recognition", "heading": "Recognized for widening access", "body": "On 21 July 2025, PKB included Data Sorcerer in its Rising Changemaker Award nominations. Its announcement highlighted access to project-based AI and data science learning. The nomination placed accessible, hands-on learning in the spotlight.", "links": [ { "label": "Read the official PKB announcement", "href": "https://pkb.id/read/12623/pkb-umumkan-nominasi-rising-changemaker-award-2025/" } ] }, { "id": "reporting", "heading": "How to read these figures", "body": "Company-reported as of 31 October 2026. Student reach describes reach across programs, not employment or completion outcomes for every participant. The grant figure is the reported value of certification and mentoring support, not cash disbursements. Program and business categories may overlap and should not be added together.", "links": [ { "label": "Ask about our impact figures", "href": "mailto:contact@data-sorcerers.com?subject=Impact%20information%20request" } ] } ] }, { "title": "Different businesses. A shared need for clarity.", "url": "https://data-sorcerers.com/customers/", "summary": "We help organizations connect their questions with the data, AI, and software needed to act. Our support spans Indonesian MSMEs and private companies, alongside medium enterprises in Singapore.", "sections": [ { "id": "starting-point", "heading": "Start with the work people actually do", "body": "A business needs a clearer view of information, a workflow that works, or a product people can use. Our first conversation looks at the current process, its users, the available data, and where the friction sits. Understanding that context comes before choosing the technology." }, { "id": "support", "heading": "Support that meets different contexts", "body": "Data Sorcerers reports supporting 15 MSMEs in Indonesia, three medium enterprises in Singapore, and five private companies in Indonesia. These are business-support counts, not a claim that every organization purchased the same service or achieved the same result. Incubation support is reported separately on our Impact page.", "links": [ { "label": "See the wider impact", "href": "/impacts/" } ] }, { "id": "delivery", "heading": "From an initial question to a shared scope", "body": "Together, we identify the users, constraints, information sources, and outcome worth pursuing. That discussion shapes a focused discovery, prototype, integration, or product build. The approach depends on the problem and the readiness of the data.", "items": [ "Intelligence platforms for monitoring, analysis, and decisions.", "Applied AI products and LLM integration for specific workflows.", "Learning and certification through Pyronyx Academy and Pyronyx Certification." ], "links": [ { "label": "Explore products and services", "href": "/products/" } ] }, { "id": "accountability", "heading": "Make progress visible", "body": "A useful delivery plan makes responsibilities, review points, and acceptance criteria clear. We discuss what success means before treating a technical metric as a business result. Documentation and handover matter because people need to understand how a solution fits their work." }, { "id": "references", "heading": "A relevant conversation before a reference", "body": "Tell us your industry and the challenge you are exploring. We can discuss relevant experience and whether a customer reference can be shared with permission. Customer identities and private project details stay with the people and organizations they belong to.", "links": [ { "label": "Discuss your project", "href": "mailto:contact@data-sorcerers.com?subject=Project%20and%20customer%20reference%20inquiry" } ] } ] }, { "title": "A place to ask. A reason to keep learning.", "url": "https://data-sorcerers.com/events/", "summary": "Our events and programs bring people together around data and AI, from a first conversation to sustained mentorship. Explore the series that form our learning community.", "sections": [ { "id": "gathering", "heading": "Sorcery Gathering · Editions 1–11", "body": "A recurring meeting point for the Data Sorcerers community. Eleven editions form part of our wider online and offline learning work, creating space for people to connect around a shared interest in data and AI. Each gathering belongs to a continuing community, rather than ending with the session.", "links": [ { "label": "Visit Data Sorcerers Community", "href": "https://community.data-sorcerers.com" } ] }, { "id": "talks", "heading": "Sorcery Talks Series", "body": "A conversation can make a technical field easier to enter. Sorcery Talks Series is part of our program mix for sharing ideas and opening discussion. Follow the community channels for individual topics, speakers, schedules, and participation details.", "links": [ { "label": "Follow program announcements", "href": "https://www.instagram.com/datasorcerers/" } ] }, { "id": "mentorship", "heading": "Sorcery AI Mentorship 2025", "body": "Mentorship gives learning more time and direction. Through the 2025 program, Data Sorcerers reports certification and mentoring grants valued at more than IDR 2.2 billion. This is the value of support through the program, rather than a cash-grant total.", "links": [ { "label": "Read the impact overview", "href": "/impacts/" } ] }, { "id": "herai", "heading": "HerAI Fellowship 2026", "body": "HerAI Fellowship 2026 adds a dedicated fellowship to our learning ecosystem. Visit HerAI for program information and community updates. Participation, dates, and application availability are communicated through that platform.", "links": [ { "label": "Visit HerAI", "href": "https://her-ai.data-sorcerers.com" } ] }, { "id": "host", "heading": "Make the next conversation relevant to your community", "body": "We welcome conversations with universities, schools, businesses, and community organizers. Share your audience, topic, preferred date and timezone, format, and what participants should take away. We will discuss fit and availability before confirming a session.", "links": [ { "label": "Propose an event", "href": "mailto:contact@data-sorcerers.com?subject=Event%20collaboration%20proposal" }, { "label": "Find current announcements", "href": "https://www.instagram.com/datasorcerers/" } ] } ] }, { "title": "Direction that stays close to people.", "url": "https://data-sorcerers.com/leadership/", "summary": "Leading Data Sorcerers means connecting learning, research, and delivery—and taking responsibility for the decisions between them.", "sections": [ { "id": "direction", "heading": "Keep the purpose in view", "body": "Technology offers many possible directions. Leadership begins by choosing which questions deserve attention and why. Our guiding idea is practical: help more people learn, build, and use data and AI in ways that matter to their work and communities." }, { "id": "disciplines", "heading": "Make room for different kinds of expertise", "body": "A researcher may see an evaluation gap. An engineer may see an operating constraint. A community organizer may see a barrier to participation. Good decisions need these perspectives together. Our leadership approach connects technical judgment with the experience of the people who use or learn from the work." }, { "id": "ownership", "heading": "Turn ambition into clear responsibilities", "body": "Our approach is to make the problem, scope, review points, and ownership explicit. Teams should understand what they can decide, where they need input, and when to raise a concern. Accountability means following through and being willing to change course when the evidence changes." }, { "id": "learning", "heading": "Build the conditions for others to grow", "body": "Our education programs, mentoring, community work, and remote internship opportunities reflect a commitment to developing capability. Leadership supports that commitment by encouraging questions, thoughtful feedback, and learning across disciplines.", "links": [ { "label": "Explore careers", "href": "https://careers.data-sorcerers.com/" }, { "label": "See our learning programs", "href": "/events/" } ] }, { "id": "conversation", "heading": "A conversation with the team", "body": "For strategic partnerships, speaking invitations, or organizational questions, contact our team with the purpose of the conversation. We will help route your inquiry to the appropriate person.", "links": [ { "label": "Contact the team", "href": "mailto:contact@data-sorcerers.com?subject=Leadership%20and%20partnership%20inquiry" } ] } ] }, { "title": "The work, and the milestones along the way.", "url": "https://data-sorcerers.com/news/", "summary": "Updates from our learning programs, business support, and community—with dates, context, and links to explore further.", "sections": [ { "id": "impact-update", "heading": "31 October 2026 · Our impact so far", "body": "Data Sorcerers reports reaching more than 10,000 school and university students across Indonesia and Southeast Asia. Collaboration spans university departments in more than 15 Indonesian cities. Our updated Impact page explains the programs, business-support figures, and reporting context.", "links": [ { "label": "Read the impact update", "href": "/impacts/" } ] }, { "id": "programs", "heading": "31 October 2026 · A growing learning ecosystem", "body": "Sorcery Gathering editions 1–11, Sorcery Talks Series, Sorcery AI Mentorship 2025, and HerAI Fellowship 2026 form part of our community learning work. The 2025 mentorship program includes reported certification and mentoring support worth more than IDR 2.2 billion.", "links": [ { "label": "Explore events and programs", "href": "/events/" } ] }, { "id": "recognition", "heading": "21 July 2025 · PKB Rising Changemaker Award nomination", "body": "PKB named Data Sorcerer among its 2025 Rising Changemaker Award nominees, highlighting access to project-based AI and data science learning. The announcement connects this nomination with the community’s approach to preparing Indonesian digital talent.", "links": [ { "label": "Read the original announcement", "href": "https://pkb.id/read/12623/pkb-umumkan-nominasi-rising-changemaker-award-2025/" } ] }, { "id": "media", "heading": "For journalists, editors, and organizers", "body": "For an interview, factual clarification, speaker request, or brand-asset inquiry, email your publication or organization, topic, deadline, and contact details. We can help identify the relevant source and contact for your story.", "links": [ { "label": "Send a media inquiry", "href": "mailto:contact@data-sorcerers.com?subject=Media%20inquiry" } ] } ] }, { "title": "What would you like to make possible?", "url": "https://data-sorcerers.com/contact-us/", "summary": "A project, a learning program, a partnership, or a question about our work. Tell us where you are starting and we will help find the right conversation.", "sections": [ { "id": "projects", "heading": "Build something useful", "body": "For a data, AI, or software project, tell us who the users are, what is difficult today, the information available, and any important timing or budget constraints. An initial conversation helps establish whether our work is a good fit.", "links": [ { "label": "Discuss a project", "href": "mailto:contact@data-sorcerers.com?subject=Project%20inquiry" }, { "label": "Message on WhatsApp", "href": "https://wa.me/62877690184444?text=Hello%20Data%20Sorcerers%2C%20I%20would%20like%20to%20discuss%20a%20project." } ] }, { "id": "learning", "heading": "Bring learning to your organization", "body": "For workshops, university collaboration, mentorship, and community programs, include the audience, topic, expected group size, preferred format, and timezone. We will discuss scope and availability with you.", "links": [ { "label": "Plan a learning collaboration", "href": "mailto:contact@data-sorcerers.com?subject=Learning%20and%20community%20collaboration" } ] }, { "id": "careers", "heading": "Apply through Careers", "body": "Our Careers portal lists all 18 open remote internships and the application form for each role. For a correction to an existing application or an accessibility request, contact us with your application reference. Please do not send identity documents or private credentials in an initial inquiry.", "links": [ { "label": "View open roles", "href": "https://careers.data-sorcerers.com/#open-positions" } ] }, { "id": "availability", "heading": "When and where to reach us", "body": "Monday–Friday, 10:00–18:00 GMT+7. Email contact@data-sorcerers.com or WhatsApp +62 877-6901-84444. Meetings are arranged in advance, including in-person conversations at our Indonesian locations.", "items": [ "Sleman, Special Region of Yogyakarta, Indonesia", "Magelang, Central Java, Indonesia", "Denpasar, Bali, Indonesia" ] }, { "id": "follow", "heading": "Stay connected between conversations", "body": "Follow Data Sorcerers for community updates and company news, or explore the communities for ways to participate.", "links": [ { "label": "Instagram", "href": "https://www.instagram.com/datasorcerers/" }, { "label": "LinkedIn", "href": "https://www.linkedin.com/company/data-sorcerers/" }, { "label": "Our communities", "href": "/community/" } ] } ] }, { "title": "Knowledge becomes useful when more people can use it.", "url": "https://data-sorcerers.com/about-us/", "summary": "We connect AI research, data intelligence, software, and learning. Our work brings people into technology—and helps organizations put that technology to work.", "sections": [ { "id": "belief", "heading": "Access and application belong together", "body": "A powerful model or a useful dataset means little if people cannot understand or use it. This idea connects our company and communities. We work on practical systems while creating opportunities for people to develop the skills, confidence, and connections to contribute." }, { "id": "community", "heading": "A community with many ways in", "body": "Someone may first encounter Data Sorcerers through a gathering, a talk, a university collaboration, or a fellowship. Our online and offline programs have reached more than 10,000 school and university students across Indonesia and Southeast Asia. Each format offers a different way to engage with data and AI.", "links": [ { "label": "Explore our communities", "href": "/community/" }, { "label": "Read our impact", "href": "/impacts/" } ] }, { "id": "work", "heading": "Technical work with a practical destination", "body": "We connect research and engineering with the decisions organizations need to make. Our ecosystem includes Data Intelligence Platform, Applied AI Products, and LLM Integration through Arutala. Pyronyx Academy and Pyronyx Certification support learning and professional development.", "links": [ { "label": "Explore our products", "href": "/products/" }, { "label": "Explore our research", "href": "/research/" } ] }, { "id": "journey", "heading": "The work continues beyond the first milestone", "body": "In 2025, our learning work included Sorcery AI Mentorship and certification and mentoring grants valued at more than IDR 2.2 billion. That year also brought a PKB Rising Changemaker Award nomination. In 2026, HerAI Fellowship and support for ten startups continue the wider story.", "links": [ { "label": "Follow our milestones", "href": "/news/" } ] }, { "id": "principles", "heading": "What we want our work to stand for", "body": "Clear thinking. Useful delivery. Openness to learning. Respect for the people and information involved. These principles guide how we frame a problem, collaborate across disciplines, evaluate a result, and communicate what we know.", "links": [ { "label": "Work with us", "href": "/contact-us/" }, { "label": "Join the team", "href": "https://careers.data-sorcerers.com/" } ] } ] }, { "title": "Resources", "url": "https://data-sorcerers.com/resources/", "summary": "Explore the resources areas available across the Data Sorcerers ecosystem.", "sections": [ { "heading": "Learn", "body": "Key resources entries you can explore now.", "items": [ "Blog", "Bootcamp", "Case Studies", "Demos", "Tutorials", "Seminar" ] }, { "heading": "Services", "body": "Key resources entries you can explore now.", "items": [ "Reports", "Updates", "Documentations", "24/7 Services", "Hire an Expert", "Professional Services" ] } ] }, { "title": "Ideas for people building with technology.", "url": "https://data-sorcerers.com/blog/", "summary": "Original analysis on AI, data, software engineering, security, and responsible technology from Data Sorcerers.", "sections": [ { "heading": "The AI demo works. The operating model is still missing.", "body": "A prototype can answer a question. A production system must fit a workflow, survive changes, and have someone accountable when it fails.", "linkSlug": "blog/ai-pilot-to-production" }, { "heading": "RAG puts your documents in the answer. It also puts them in the critical path.", "body": "Retrieval can connect an AI assistant to organizational knowledge. Its usefulness depends on what is retrieved, who may see it, and whether it is still current.", "linkSlug": "blog/rag-knowledge-quality" }, { "heading": "Before adding an AI agent, draw the boundary around its work.", "body": "Autonomy is an architectural choice. Teams need to decide where a system may choose its next step—and where a person or fixed rule must remain in control.", "linkSlug": "blog/ai-agents-workflow-boundaries" }, { "heading": "The case for a smaller model begins with the task.", "body": "Model size is a constraint to evaluate, not a shortcut to quality. Narrow workloads can change the trade-off between capability, latency, and deployment complexity.", "linkSlug": "blog/small-language-models-deployment" }, { "heading": "A good model score can still hide a bad product decision.", "body": "Evaluation needs to connect an aggregate metric with the people, situations, and errors that matter in use.", "linkSlug": "blog/model-evaluation-beyond-accuracy" }, { "heading": "Data quality is a daily operating decision.", "body": "A validation tool can detect an unexpected value. Someone still needs to decide what that value means and what happens next.", "linkSlug": "blog/data-quality-as-an-operating-practice" }, { "heading": "The dashboard disagreement may be a definition problem.", "body": "Before redesigning business intelligence, establish what the numbers mean, when they are measured, and who can change their definitions.", "linkSlug": "blog/business-intelligence-shared-definitions" }, { "heading": "A forecast should face the future before it faces the boardroom.", "body": "Testing with information that would not have been available at prediction time can make a forecasting system look more capable than it is.", "linkSlug": "blog/forecasting-without-future-leakage" }, { "heading": "A satellite image is an observation, not a complete explanation.", "body": "Geospatial intelligence becomes useful when location, timing, resolution, and ground context are part of the interpretation.", "linkSlug": "blog/satellite-data-decision-context" }, { "heading": "A smaller architecture can leave more room to grow.", "body": "Splitting software into services is a deployment decision. Separating responsibilities is a design discipline that matters at every scale.", "linkSlug": "blog/modular-software-before-microservices" }, { "heading": "An API is a promise to the team on the other side.", "body": "Clear interfaces help systems evolve independently. The difficult work is deciding which promises must remain stable when the implementation changes.", "linkSlug": "blog/api-contracts-for-change" }, { "heading": "Uptime is not the same as a working user journey.", "body": "A service can answer requests while the task that matters to the user still fails. Reliability measures need to follow the experience.", "linkSlug": "blog/reliability-user-journeys" }, { "heading": "When an AI assistant reads the web, it crosses a trust boundary.", "body": "Retrieved material can contain instructions as well as information. The system must keep outside content separate from authority to act.", "linkSlug": "blog/prompt-injection-trust-boundaries" }, { "heading": "A fast interface is only useful if people can operate it.", "body": "Performance and accessibility meet in the same user journey: finding information, understanding controls, and completing a task without unnecessary friction.", "linkSlug": "blog/performance-accessibility-shared-design" }, { "heading": "Responsible AI becomes real in ordinary product decisions.", "body": "Governance matters when it changes what a team builds, how it evaluates the result, and who can intervene when the system causes problems.", "linkSlug": "blog/responsible-ai-operating-decisions" } ] }, { "title": "Core AI", "url": "https://data-sorcerers.com/core-ai/", "summary": "Teach computers to learn from examples, find patterns, and make better predictions.", "sections": [ { "heading": "What we study", "body": "Teach computers to learn from examples, find patterns, and make better predictions.", "items": [ "Machine Learning", "Deep Learning", "Reinforcement Learning" ] } ] }, { "title": "Language & Reasoning", "url": "https://data-sorcerers.com/language-and-reasoning/", "summary": "Help AI understand documents, answer questions using your information, and support everyday tasks.", "sections": [ { "heading": "What we study", "body": "Help AI understand documents, answer questions using your information, and support everyday tasks.", "items": [ "Natural Language Processing", "Large Language Models", "Knowledge Graphs", "Agentic AI" ] } ] }, { "title": "Vision & Multimodal", "url": "https://data-sorcerers.com/vision-and-multimodal/", "summary": "Help AI understand images, video, and text together—for tasks such as reading documents or inspecting objects.", "sections": [ { "heading": "What we study", "body": "Help AI understand images, video, and text together—for tasks such as reading documents or inspecting objects.", "items": [ "Computer Vision", "Multimodal AI", "Generative AI", "Remote Sensing AI" ] } ] }, { "title": "Data Intelligence", "url": "https://data-sorcerers.com/data-intelligence/", "summary": "Turn scattered data into useful insights, forecasts, and early warnings your team can act on.", "sections": [ { "heading": "What we study", "body": "Turn scattered data into useful insights, forecasts, and early warnings your team can act on.", "items": [ "Data Mining", "Time Series Forecasting", "Recommender Systems", "Anomaly Detection" ] } ] }, { "title": "AI Engineering", "url": "https://data-sorcerers.com/ai-engineering/", "summary": "Turn AI prototypes into reliable applications, then monitor and improve them as people use them.", "sections": [ { "heading": "What we study", "body": "Turn AI prototypes into reliable applications, then monitor and improve them as people use them.", "items": [ "MLOps", "Model Deployment", "AI Evaluation", "LLMOps" ] } ] }, { "title": "Trustworthy AI", "url": "https://data-sorcerers.com/trustworthy-ai/", "summary": "Test AI for mistakes and unfair results, protect private data, and keep people in control of important decisions.", "sections": [ { "heading": "What we study", "body": "Test AI for mistakes and unfair results, protect private data, and keep people in control of important decisions.", "items": [ "AI Safety", "Responsible AI", "Explainable AI", "Privacy-Preserving AI" ] } ] }, { "title": "Privacy Policy", "url": "https://data-sorcerers.com/privacy-policy/", "summary": "How Data Sorcerers handles information from this company website and your inquiries, and how to exercise your privacy rights.", "sections": [ { "id": "scope", "heading": "01 / Scope and responsibility", "body": "PT. Data Sorcerers Indonesia operates this company website. This notice covers visitors and people who contact us about services, careers, partnerships, or other inquiries. Separate products, including Pyronyx Certification, have their own notices for accounts, assessments, payments, and other product activities. Visiting this website does not enroll you in those services." }, { "id": "information", "heading": "02 / Information we handle", "body": "The information involved depends on how you interact with us. This website has no account registration, payment checkout, file-upload form, or exam recording feature.", "items": [ "When you email us: your email address, name if provided, message, attachments, and details you choose to share about your organization or request.", "When your browser loads a page: technical connection information, such as the IP address, requested resource and browser request headers, is processed to deliver the website. Hosting services may retain operational or security logs under their own arrangements.", "This company website does not load advertising pixels or request camera, microphone, or location access. Fonts, imagery, and community website previews are served as local assets; previews are screenshots, not embedded third-party websites." ] }, { "id": "assistant", "heading": "03 / Website assistant and contact links", "body": "The assistant is a local knowledge demo, not a live AI model. Questions and replies are processed in this browser tab and are not sent to an AI provider or saved in a website account. Reset chat clears the conversation; reloading or closing the tab also clears it. Avoid including confidential or sensitive information. A WhatsApp link opens WhatsApp with an editable message, and an email link opens your mail application. We receive the message only when you send it through that service." }, { "id": "purposes", "heading": "04 / Purposes and processing bases", "body": "We use inquiry information to respond, discuss a requested service, arrange follow-up, and handle support or disputes. Relevant information may be needed for steps you request before an agreement, a contractual obligation, a legal duty, or a proportionate legitimate interest in communication and security. Where processing depends on consent, it must be specific to that purpose. Sending an inquiry is not a subscription to marketing." }, { "id": "sharing", "heading": "05 / Sharing and external services", "body": "Information may be handled by hosting and email providers needed to deliver pages or correspondence, by authorized staff handling your request, or by professional advisers and authorities where necessary and lawful. We do not sell personal data collected through this website. External links open separate services governed by their own notices; simply displaying a link does not embed their tracking tools. Contact us for the provider and location arrangements relevant to your inquiry. We do not promise that every communication is stored only in Indonesia." }, { "id": "hosting", "heading": "06 / Hosting, security, and service providers", "body": "Our website is delivered through Cloudflare and a hosted server. These services process technical request information, including IP addresses and browser headers, to deliver content, prevent abuse, and diagnose errors. Cloudflare may also collect cookieless performance measurements such as page-load timing and page URL through its Web Analytics beacon. This is separate from advertising or a marketing profile. Cloudflare may set security or challenge cookies depending on the protections applied to a request. Email, WhatsApp, Instagram, LinkedIn, and linked platforms have their own privacy practices when you use them.", "links": [ { "label": "Cloudflare privacy notice", "href": "https://www.cloudflare.com/privacypolicy/" }, { "label": "Cloudflare cookies", "href": "https://developers.cloudflare.com/fundamentals/reference/policies-compliances/cloudflare-cookies/" } ] }, { "id": "retention", "heading": "07 / Retention and deletion", "body": "Inquiry correspondence is handled through our communication channels rather than a website account database. This website does not set an automatic deletion period for email, provider logs, or backups. Retention depends on the purpose, the relevant provider arrangements, ongoing requests or disputes, and applicable recordkeeping duties. You can ask about the retention arrangement for your information or request deletion; any necessary limits and backup implications should be explained in the response." }, { "id": "rights", "heading": "08 / Your choices and rights", "body": "Subject to applicable law, you may ask for information about processing, access, correction, deletion, restriction, withdrawal of consent, a usable copy, or review of a decision based solely on automated processing. This company website does not make automated eligibility decisions. Applicable legal response periods govern requests.", "items": [ "Email contact@data-sorcerers.com with the subject of your request and enough context to identify the relevant correspondence.", "We may ask for proportionate information to verify a request. Do not send passwords, access tokens, or unnecessary identity documents.", "Withdrawal does not invalidate earlier lawful processing. If a request cannot be fully met, you can ask for the reason and available next steps.", "You may raise unresolved concerns through complaint or dispute procedures available under applicable law." ], "links": [ { "label": "Indonesia’s Personal Data Protection Law", "href": "https://jdih.komdigi.go.id/produk_hukum/view/id/832/t/crc32/" }, { "label": "Contact our team", "href": "mailto:contact@data-sorcerers.com" } ] }, { "id": "security", "heading": "09 / Security and information you send", "body": "Send only what is needed for your inquiry. Before sharing confidential datasets, credentials, or sensitive personal information for a project, agree on an appropriate transfer channel and the relevant safeguards with our team. No website or email channel can guarantee absolute security. Suspected exposure or misuse can be reported to our contact address; personal-data incidents must be handled according to applicable law." }, { "id": "children", "heading": "10 / Children and separate services", "body": "This company website presents business information and does not provide child accounts or knowingly request children’s personal data. If you believe a child has sent personal information through an inquiry, contact us so the request can be reviewed. Learning and certification services have their own participation and privacy notices." }, { "id": "updates", "heading": "11 / Contact and updates", "body": "The version and update date identify this notice. Changes in website features or processing should be reflected here before they apply. For privacy questions, requests, or concerns, contact PT. Data Sorcerers Indonesia at contact@data-sorcerers.com.", "links": [ { "label": "Contact our team", "href": "mailto:contact@data-sorcerers.com" } ] } ] }, { "title": "Terms & Conditions", "url": "https://data-sorcerers.com/terms-condition/", "summary": "The terms for using the Data Sorcerers company website, its materials, and its contact channels.", "sections": [ { "id": "scope", "heading": "01 / Operator and scope", "body": "This website is operated by PT. Data Sorcerers Indonesia and provides information about our work, capabilities, research, and related services. These terms apply to use of this company website. Separate products and programs have their own participation rules. Visiting the website does not create a paid engagement, purchase, enrollment, or partnership." }, { "id": "use", "heading": "02 / Responsible use", "body": "Use the website lawfully and respect other people’s information and rights.", "items": [ "Do not attempt unauthorized access, interfere with availability, introduce malicious code, or bypass security controls.", "Do not impersonate another person, send unlawful material, or use contact details for spam.", "Do not misrepresent an affiliation, certification, endorsement, or agreement with Data Sorcerers.", "Report suspected security issues privately without disclosing personal data or exploiting the issue beyond what is necessary to describe it." ] }, { "id": "engagements", "heading": "03 / Proposals and paid engagements", "body": "A service description or inquiry does not confirm scope, price, availability, or delivery dates. A paid engagement requires a separate agreement defining deliverables, responsibilities, fees, acceptance, confidentiality, data handling, and ownership. That agreement governs the engagement, subject to mandatory law. Scope changes may require an agreed adjustment to fees or timelines." }, { "id": "materials", "heading": "04 / Content and intellectual property", "body": "Website content, branding, illustrations, research materials, and software belong to Data Sorcerers or their respective rights holders. You may link to public pages and quote brief, attributed excerpts for a lawful informational purpose. Substantial republication, commercial reuse of visual assets, or use of our brand implying endorsement requires permission. Rights in project deliverables and client materials are determined by the applicable agreement." }, { "id": "inquiries", "heading": "05 / Information you provide", "body": "Provide accurate contact information and share only what is needed to discuss your request. Do not send passwords, access keys, third-party personal data, or confidential project files through an initial inquiry. A confidentiality or data-processing arrangement should be established before sensitive project information is transferred. Privacy handling is described in the Privacy Policy." }, { "id": "information", "heading": "06 / Website information and external links", "body": "Materials describe our services and research for general information. They are not individualized legal, financial, medical, investment, or other professional advice, and do not guarantee a business outcome. AI and analytical results require evaluation in the relevant context. External links lead to separate services; those services control their own content and terms. The website assistant is a scripted knowledge guide; its replies do not constitute a quote, a contract, or a commitment to a delivery date. Confirm project-specific details with our team." }, { "id": "availability", "heading": "07 / Availability and legal rights", "body": "Website access can be affected by maintenance, network problems, or third-party service interruptions. We do not promise uninterrupted access or that every informational item is free of errors. Responsibility for any loss is determined by the relevant agreement and applicable law. Nothing in these terms excludes liability or removes consumer, personal-data, or other rights that cannot lawfully be excluded." }, { "id": "disputes", "heading": "08 / Governing law and concerns", "body": "These terms are governed by the laws of the Republic of Indonesia. Contact us first with the relevant page and a clear description of your concern so it can be reviewed. Unresolved matters remain subject to applicable complaint and dispute-resolution procedures. These terms do not impose compulsory arbitration or waive statutory remedies.", "links": [ { "label": "Contact our team", "href": "mailto:contact@data-sorcerers.com" } ] }, { "id": "updates", "heading": "09 / Changes and contact", "body": "The version and update date identify these terms. Updates should explain material changes to website use and do not silently amend an existing signed project agreement. If a provision cannot be enforced, applicable law determines its effect on the remaining terms. Contact contact@data-sorcerers.com for website concerns or permissions.", "links": [ { "label": "Contact our team", "href": "mailto:contact@data-sorcerers.com" } ] } ] }, { "title": "Copyright", "url": "https://data-sorcerers.com/copyright/", "summary": "Copyright, brand, content, software, research, visual asset, and permitted-use information for Data Sorcerers materials.", "sections": [ { "heading": "Ownership", "body": "Unless otherwise stated, Data Sorcerers owns or licenses the content, brand elements, interface concepts, visual assets, research summaries, copy, and materials presented on this website.", "items": [ "Data Sorcerers name, marks, UI direction, page copy, and brand expressions", "Original website visuals, generated or commissioned imagery, diagrams, and layouts", "Reusable methods, templates, software concepts, research descriptions, and documentation" ] }, { "heading": "Permitted use", "body": "You may quote short excerpts or reference Data Sorcerers factually for media, procurement, partnership, or evaluation purposes, provided the use is accurate, attributed, and not misleading.", "items": [ "Do not imply endorsement, partnership, or certification without written permission", "Do not reuse substantial content, visuals, generated images, or interface designs as your own", "Do not modify brand assets in a way that misrepresents Data Sorcerers" ] }, { "heading": "Third-party materials", "body": "Some tools, fonts, libraries, infrastructure, or illustrative assets may be provided by third parties and remain subject to their own licenses or terms." }, { "heading": "Permissions and takedowns", "body": "For licensing, republication, brand usage, image permissions, attribution corrections, or copyright concerns, contact contact@data-sorcerers.com with the relevant URL, material, and requested action." } ] }, { "title": "The AI demo works. The operating model is still missing.", "url": "https://data-sorcerers.com/blog/ai-pilot-to-production/", "summary": "A prototype can answer a question. A production system must fit a workflow, survive changes, and have someone accountable when it fails.", "sections": [ { "id": "section-1", "heading": "The difficult part starts after the demonstration", "body": "An AI prototype can make a familiar task look effortless. A document becomes a summary; a spreadsheet becomes an answer. The demonstration proves that a capability exists. It does not yet establish who will review the output, which information may enter the system, or how the organization will handle an incorrect answer. Those decisions determine whether the prototype becomes useful work or another isolated experiment." }, { "id": "section-2", "heading": "A model is one dependency among many", "body": "The research paper Hidden Technical Debt in Machine Learning Systems describes risks that extend beyond model code, including data dependencies, feedback loops, configuration, and changes in the surrounding environment. Its practical relevance remains architectural: a prediction is produced by a system with many moving parts. Improving the model does not automatically improve the workflow around it.", "links": [ { "label": "Sculley et al. — Hidden Technical Debt in Machine Learning Systems (2015)", "href": "https://papers.nips.cc/paper_files/paper/2015/hash/86df7dcfd896fcaf2674f757a2463eba-Abstract.html" } ] }, { "id": "section-3", "heading": "Write the operating brief before the next experiment", "body": "Consider a hypothetical support team testing an assistant for internal policy questions. The team should define which policies are authoritative, who owns revisions, and which answers require escalation. It should also decide whether staff are looking for a quick pointer or a response they can send to a customer. These are different products, even if they use the same model. A brief that names the user, decision, reviewer, and failure path gives the next experiment a concrete purpose." }, { "id": "section-4", "heading": "Measure the work that remains", "body": "Our recommendation is to compare the full task with the current process. Include time spent locating context, checking the answer, correcting it, and documenting the result. Keep the comparison on similar tasks. If an assistant saves drafting time but increases verification effort, the apparent gain may disappear. Report quality and workload together rather than presenting the fastest successful example as typical performance." }, { "id": "section-5", "heading": "Make ownership part of the release", "body": "A practical release review should identify an operational owner, a way to disable the feature, and a plan for reviewing failures. Record which model, prompt, data sources, and permissions were tested. Start with a limited workflow whose consequences are understood. Expansion should follow evidence from use, not the excitement of a successful demonstration. The useful milestone is a team that can operate the system with confidence." }, { "heading": "The practical takeaway", "body": "Before expanding an AI pilot, name the workflow owner, the review path, and the outcome you will measure." }, { "heading": "Sources & further reading", "body": "Primary sources consulted for this analysis. Examples are hypothetical; recommendations are our editorial interpretation.", "links": [ { "label": "Sculley et al. — Hidden Technical Debt in Machine Learning Systems (2015)", "href": "https://papers.nips.cc/paper_files/paper/2015/hash/86df7dcfd896fcaf2674f757a2463eba-Abstract.html" } ] } ] }, { "title": "RAG puts your documents in the answer. It also puts them in the critical path.", "url": "https://data-sorcerers.com/blog/rag-knowledge-quality/", "summary": "Retrieval can connect an AI assistant to organizational knowledge. Its usefulness depends on what is retrieved, who may see it, and whether it is still current.", "sections": [ { "id": "section-1", "heading": "The knowledge problem is not just a model problem", "body": "An internal assistant is often asked to explain information that changes faster than its underlying model. That makes retrieval attractive: the system can look up relevant material before generating an answer. But a folder of documents is not automatically a dependable knowledge base. Old procedures, duplicate drafts, and conflicting instructions can all enter the same search result unless someone decides what belongs there." }, { "id": "section-2", "heading": "What the original research established", "body": "The 2020 retrieval-augmented generation paper combined a pretrained language model with a retrievable external collection. Its experiments explored knowledge-intensive language tasks and showed advantages over the tested baselines. That research supports a design pattern, not a guarantee for every corporate assistant. A deployment still needs evidence that its own retrieval and answers work for its users.", "links": [ { "label": "Lewis et al. — Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks", "href": "https://arxiv.org/abs/2005.11401" } ] }, { "id": "section-3", "heading": "Build a collection people can govern", "body": "For a hypothetical employee handbook assistant, start with approved policies rather than every file in a shared drive. Attach an owner, effective date, and document identifier to each source. Keep permissions aligned with the people asking questions. Decide how archived material is removed and how revised documents replace earlier versions. This work may be less visible than the chat interface, but it determines what evidence the assistant can use." }, { "id": "section-4", "heading": "Test retrieval separately from writing", "body": "Our suggested evaluation has two stages. First, check whether the system retrieves the material needed for a question. Then examine whether the answer accurately represents that material and handles missing information. A fluent answer can conceal a retrieval failure. A good passage can also be summarized incorrectly. Keeping these checks separate makes it easier to identify which part needs improvement." }, { "id": "section-5", "heading": "Design for disagreement and absence", "body": "Try questions with no answer in the collection, conflicting policies, and references to an obsolete procedure. Decide what the user should see in each case: an explicit limitation, competing sources, or a route to a person. A citation is useful only when it supports the associated statement and the user is entitled to inspect it. Treat source access and maintenance as continuing responsibilities, not one-time ingestion tasks." }, { "heading": "The practical takeaway", "body": "Evaluate the evidence path as carefully as the generated answer." }, { "heading": "Sources & further reading", "body": "Primary sources consulted for this analysis. Examples are hypothetical; recommendations are our editorial interpretation.", "links": [ { "label": "Lewis et al. — Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks", "href": "https://arxiv.org/abs/2005.11401" } ] } ] }, { "title": "Before adding an AI agent, draw the boundary around its work.", "url": "https://data-sorcerers.com/blog/ai-agents-workflow-boundaries/", "summary": "Autonomy is an architectural choice. Teams need to decide where a system may choose its next step—and where a person or fixed rule must remain in control.", "sections": [ { "id": "section-1", "heading": "A useful question before the technology decision", "body": "When a process is slow, an agent can sound like a complete answer. Yet many tasks are slow because their inputs are unclear, approvals are duplicated, or information is scattered. Automating those conditions may make the confusion move faster. Begin by drawing the process as it exists today. Identify which steps are predictable and which require interpretation before deciding how much autonomy the system needs." }, { "id": "section-2", "heading": "Separate a workflow from an agent", "body": "Anthropic’s 2024 engineering guidance distinguishes predefined workflows from agents that dynamically direct their process and tool use. It argues for increasing complexity only when needed. We use that distinction as a design lens: a fixed sequence can be appropriate when the route is known, while dynamic planning may help where the route depends on what the system discovers.", "links": [ { "label": "Anthropic — Building effective agents (2024)", "href": "https://www.anthropic.com/research/building-effective-agents" } ] }, { "id": "section-3", "heading": "Give the system a bounded assignment", "body": "Consider a hypothetical research assistant collecting public information about suppliers. It could search approved sources, organize evidence, and draft a briefing. That does not imply permission to contact a supplier, change a purchase order, or send confidential documents. Describe permitted actions at the tool and application layers. A prompt that says “be careful” should not be the only thing separating a research task from an external commitment." }, { "id": "section-4", "heading": "Observe the route, not only the destination", "body": "Our recommended evaluation records the task, tool calls, important intermediate decisions, costs, and final result. Review unnecessary loops as well as obvious failures. An apparently correct answer may have required excessive browsing or relied on a source that should not have been accessible. Give the operator a clear stop action and make repeated failures visible. The system should be inspectable without reading every internal token." }, { "id": "section-5", "heading": "Expand authority deliberately", "body": "A sensible first deployment might draft a recommendation while a person approves the next step. If evidence later supports more automation, expand a specific permission rather than granting broad control. Define time, budget, and action limits for each assignment. The point is not to maximize the number of autonomous steps. It is to complete useful work within boundaries that the organization can explain and supervise." }, { "heading": "The practical takeaway", "body": "Match autonomy to the task’s uncertainty and consequences, not to the appeal of the word “agent.”" }, { "heading": "Sources & further reading", "body": "Primary sources consulted for this analysis. Examples are hypothetical; recommendations are our editorial interpretation.", "links": [ { "label": "Anthropic — Building effective agents (2024)", "href": "https://www.anthropic.com/research/building-effective-agents" } ] } ] }, { "title": "The case for a smaller model begins with the task.", "url": "https://data-sorcerers.com/blog/small-language-models-deployment/", "summary": "Model size is a constraint to evaluate, not a shortcut to quality. Narrow workloads can change the trade-off between capability, latency, and deployment complexity.", "sections": [ { "id": "section-1", "heading": "Start with the workload you can describe", "body": "A team choosing a language model is often shown broad benchmark rankings. Its actual workload may be much narrower: extracting a few fields, routing an inquiry, or drafting a short response from approved material. The right comparison begins with those tasks. A model that is impressive across many domains may be unnecessary for one bounded workflow, while a smaller model may fail on an edge case the business cannot ignore." }, { "id": "section-2", "heading": "What research shows—and what it leaves to you", "body": "The Phi-3 technical report introduced a compact language model designed to deliver substantial capability at a smaller scale, including phone deployment. It is a useful example of research into efficient models. Its reported benchmarks describe the authors’ experiments; they are not evidence that an arbitrary smaller model will satisfy a particular organization’s requirements.", "links": [ { "label": "Abdin et al. — Phi-3 Technical Report (2024)", "href": "https://arxiv.org/abs/2404.14219" } ] }, { "id": "section-3", "heading": "Build a task-specific comparison", "body": "For a hypothetical document-routing service, assemble examples from the actual document formats and languages it will encounter. Include unclear scans, missing fields, and documents that belong to none of the supported categories. Compare task quality, response time, operating cost, and failure handling. Keep the surrounding prompt and retrieval setup documented, because changing those can alter the result as much as changing the model." }, { "id": "section-4", "heading": "Deployment changes the calculation", "body": "Our recommendation is to examine the full operating environment. Local inference may reduce dependency on a remote endpoint, but it introduces responsibilities for device resources, updates, and monitoring. Hosted inference moves some infrastructure work elsewhere while adding service and connectivity considerations. Data sensitivity should shape the design, but local deployment alone does not resolve access control, retention, or logging decisions." }, { "id": "section-5", "heading": "Keep a route for difficult cases", "body": "A production design can separate straightforward tasks from requests that need additional review. If a model is uncertain or receives an unsupported input, the system should have a defined response rather than producing a confident guess. Measure how often that route is used and what it costs. A smaller model is valuable when the complete workflow meets its objectives, not merely when a demonstration runs on less hardware." }, { "heading": "The practical takeaway", "body": "Compare complete workflows on your own tasks before choosing a model by size." }, { "heading": "Sources & further reading", "body": "Primary sources consulted for this analysis. Examples are hypothetical; recommendations are our editorial interpretation.", "links": [ { "label": "Abdin et al. — Phi-3 Technical Report (2024)", "href": "https://arxiv.org/abs/2404.14219" } ] } ] }, { "title": "A good model score can still hide a bad product decision.", "url": "https://data-sorcerers.com/blog/model-evaluation-beyond-accuracy/", "summary": "Evaluation needs to connect an aggregate metric with the people, situations, and errors that matter in use.", "sections": [ { "id": "section-1", "heading": "One number cannot describe every failure", "body": "A model evaluation often reaches a meeting as a single percentage. That number can be useful, but it compresses many different situations into one result. Before using it to approve a release, ask which examples were tested and which errors it conceals. A failure on a common low-consequence task and a failure on a rare critical task should not automatically carry the same meaning." }, { "id": "section-2", "heading": "Keep unfamiliar data genuinely unfamiliar", "body": "Google’s machine-learning guidance describes overfitting as strong performance on training data that does not carry over to new examples. It highlights the importance of generalization and suitable data partitions. The practical implication is straightforward: an evaluation should test behavior beyond the examples used to shape the model, rather than repeatedly rewarding familiarity with the same material.", "links": [ { "label": "Google Machine Learning Crash Course — Overfitting", "href": "https://developers.google.com/machine-learning/crash-course/overfitting/overfitting" } ] }, { "id": "section-3", "heading": "Design a review around the use case", "body": "Consider a hypothetical system that assigns incoming messages to service teams. Its aggregate accuracy could hide weak results for a language, a new product, or messages with multiple requests. Review these groups separately. Read actual failures with the people who receive the routed work. They may identify a distinction that the current labels do not capture, or show that a technically incorrect label is operationally harmless." }, { "id": "section-4", "heading": "Agree on the release decision first", "body": "Our recommendation is to decide what evidence would justify launch before comparing model candidates. Name unacceptable failure types, the review workload the team can absorb, and the situations that require a fallback. Keep a simple baseline in the comparison. If the evaluation criteria change after seeing results, document why and test again on material that was not used to make that change." }, { "id": "section-5", "heading": "Continue evaluation after launch", "body": "Real use will introduce new wording, new sources, and unexpected combinations. Capture issues in a reviewable way while respecting privacy and access requirements. Add representative failures to future evaluation sets, and retain a separate test for independent checks. The aim is a repeatable release decision, not a score that can only be reproduced by the person who built the first experiment." }, { "heading": "The practical takeaway", "body": "Use aggregate scores to start a discussion, then inspect the errors that could change the decision." }, { "heading": "Sources & further reading", "body": "Primary sources consulted for this analysis. Examples are hypothetical; recommendations are our editorial interpretation.", "links": [ { "label": "Google Machine Learning Crash Course — Overfitting", "href": "https://developers.google.com/machine-learning/crash-course/overfitting/overfitting" } ] } ] }, { "title": "Data quality is a daily operating decision.", "url": "https://data-sorcerers.com/blog/data-quality-as-an-operating-practice/", "summary": "A validation tool can detect an unexpected value. Someone still needs to decide what that value means and what happens next.", "sections": [ { "id": "section-1", "heading": "The problem usually arrives as a business question", "body": "A report changes sharply overnight. Two teams disagree about the customer count. A dashboard contains a blank where a decision should be. These incidents are often described as data problems, but resolving them requires more than a corrected query. Someone must explain what the data represents, whether the change is legitimate, and which downstream decisions may have been affected." }, { "id": "section-2", "heading": "Turn assumptions into explicit checks", "body": "Great Expectations documents a workflow for defining expectations about data and validating it against those expectations. This provides a concrete way to make assumptions testable. A check can help surface a mismatch; it cannot independently determine the business meaning of every mismatch. The team still needs an owner and an agreed response.", "links": [ { "label": "Great Expectations — GX Core overview", "href": "https://docs.greatexpectations.io/docs/core/introduction/gx_overview/" } ] }, { "id": "section-3", "heading": "Start with a critical decision", "body": "For a hypothetical admissions dashboard, a missing applicant identifier may prevent reliable counting, while a missing optional comment may not. Begin with the fields required for the decision. Document definitions, acceptable values, timing, and the source of truth. Resist adding every imaginable check before anyone has agreed which failures matter. A smaller set of meaningful checks is easier to operate than a large collection of unexplained alerts." }, { "id": "section-4", "heading": "Make the response part of the check", "body": "Our recommendation is to attach a consequence to each important validation rule. Should a failed import stop publication, flag a warning, or continue with an explicit limitation? Who investigates, and what information do they need? Record the affected dataset and the time window. This makes it possible to distinguish a genuine change in the world from a broken ingestion step without relying on informal memory." }, { "id": "section-5", "heading": "Review the rules as the business changes", "body": "A new product, enrollment cycle, or reporting definition can make an old expectation obsolete. Schedule a review of both noisy alerts and silent gaps. Keep a record of why a rule changed and who approved it. Data quality becomes dependable when definitions, checks, and responses evolve together. The goal is not a dashboard with no warnings; it is a team that understands what the warnings mean." }, { "heading": "The practical takeaway", "body": "Every important data check needs an owner, a business meaning, and a response." }, { "heading": "Sources & further reading", "body": "Primary sources consulted for this analysis. Examples are hypothetical; recommendations are our editorial interpretation.", "links": [ { "label": "Great Expectations — GX Core overview", "href": "https://docs.greatexpectations.io/docs/core/introduction/gx_overview/" } ] } ] }, { "title": "The dashboard disagreement may be a definition problem.", "url": "https://data-sorcerers.com/blog/business-intelligence-shared-definitions/", "summary": "Before redesigning business intelligence, establish what the numbers mean, when they are measured, and who can change their definitions.", "sections": [ { "id": "section-1", "heading": "Two correct queries can answer different questions", "body": "One team counts active customers as people who logged in. Another counts customers with a completed purchase. Both can build a technically correct dashboard and still disagree in a planning meeting. The conflict is not necessarily a visualization problem. It is a disagreement about the question being measured, hidden behind a familiar label." }, { "id": "section-2", "heading": "Central definitions can reduce repeated interpretation", "body": "The dbt Semantic Layer documentation describes defining metrics centrally and using those definitions across downstream tools. It presents a way to move repeated business calculations into a shared modeling layer. The broader lesson is about ownership: a metric should not acquire a different meaning every time it is copied into another report.", "links": [ { "label": "dbt Developer Hub — dbt Semantic Layer", "href": "https://docs.getdbt.com/docs/use-dbt-semantic-layer/dbt-sl" } ] }, { "id": "section-3", "heading": "Write a useful metric specification", "body": "For a hypothetical subscription business, define what counts as a subscriber, how trials are treated, and which timezone closes the reporting day. Specify whether canceled subscriptions remain in a period-end snapshot. Include the intended business question and examples of records that should be included or excluded. A definition written only as SQL can be difficult for the decision-maker to challenge." }, { "id": "section-4", "heading": "Treat a change as a product change", "body": "Our recommendation is to version important metric definitions and explain their effect on historical comparisons. If a team changes the meaning of an active account, decide whether to recalculate earlier periods or display a break in the series. Give dependent reports a clear migration path. A silent change can make a stable business look volatile—or make a genuine problem disappear." }, { "id": "section-5", "heading": "Design the screen after the decision is clear", "body": "Once the meaning is agreed, choose a view that helps someone act. Show the relevant period, comparison, update time, and limitations. Give users a route to inspect the definition rather than expecting them to remember it. Business intelligence improves when the conversation moves from “whose number is right?” to “what does this shared measure tell us to investigate?”" }, { "heading": "The practical takeaway", "body": "Agree on the meaning of a metric before optimizing its chart." }, { "heading": "Sources & further reading", "body": "Primary sources consulted for this analysis. Examples are hypothetical; recommendations are our editorial interpretation.", "links": [ { "label": "dbt Developer Hub — dbt Semantic Layer", "href": "https://docs.getdbt.com/docs/use-dbt-semantic-layer/dbt-sl" } ] } ] }, { "title": "A forecast should face the future before it faces the boardroom.", "url": "https://data-sorcerers.com/blog/forecasting-without-future-leakage/", "summary": "Testing with information that would not have been available at prediction time can make a forecasting system look more capable than it is.", "sections": [ { "id": "section-1", "heading": "The clock is part of the dataset", "body": "Forecasting is different from explaining a completed period. A model must make a useful estimate with only the information available at a particular moment. A column can look harmless in a spreadsheet yet contain knowledge from after that moment. If it enters evaluation, the model receives an advantage that will not exist when the team actually needs a forecast." }, { "id": "section-2", "heading": "Preserve the order of information", "body": "The scikit-learn TimeSeriesSplit documentation describes cross-validation for time-ordered data, where training sets precede later test sets. It also explains that ordinary cross-validation methods can train on future data and evaluate on the past. This is a useful starting point for evaluation design, though the appropriate split and gap still depend on the application.", "links": [ { "label": "scikit-learn — TimeSeriesSplit documentation", "href": "https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.TimeSeriesSplit.html" } ] }, { "id": "section-3", "heading": "Recreate the decision date", "body": "Consider a hypothetical retailer planning next week’s inventory. Ask when sales, returns, promotions, and stock availability became known. A finalized monthly report may contain corrections that were unavailable during the original planning meeting. Reconstructing that information boundary is more important than choosing an impressive model. Keep the forecast horizon explicit: predicting tomorrow and predicting the next quarter are different tasks." }, { "id": "section-4", "heading": "Compare with a baseline people understand", "body": "Our recommendation is to compare the proposed system with a simple operational baseline, such as the previous comparable period, where appropriate. Examine several historical cutoffs rather than one favorable interval. Discuss which errors are costly: over-ordering, under-ordering, or responding too late. A lower average error does not automatically produce a better inventory decision if the remaining mistakes occur at the worst moments." }, { "id": "section-5", "heading": "Show uncertainty in the operating plan", "body": "A forecast should support a decision under uncertainty, not remove uncertainty from the conversation. Explain what information it uses, where it performs poorly, and when a person should override or investigate it. Keep predictions issued at each cutoff so they can be compared with what subsequently happened. A credible forecasting process is one the team can audit without reconstructing its history from memory." }, { "heading": "The practical takeaway", "body": "Evaluate a forecast using the information that existed when the decision would have been made." }, { "heading": "Sources & further reading", "body": "Primary sources consulted for this analysis. Examples are hypothetical; recommendations are our editorial interpretation.", "links": [ { "label": "scikit-learn — TimeSeriesSplit documentation", "href": "https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.TimeSeriesSplit.html" } ] } ] }, { "title": "A satellite image is an observation, not a complete explanation.", "url": "https://data-sorcerers.com/blog/satellite-data-decision-context/", "summary": "Geospatial intelligence becomes useful when location, timing, resolution, and ground context are part of the interpretation.", "sections": [ { "id": "section-1", "heading": "A striking map can invite a premature conclusion", "body": "A colored patch on a map can look authoritative. Yet a change in an image does not, by itself, explain why the change happened or what a team should do. Before turning an observation into an operational claim, identify the area, the dates, the conditions of collection, and the decision the map is meant to support. Those details belong with the visual, not in a forgotten appendix." }, { "id": "section-2", "heading": "Know what the instrument measures", "body": "Copernicus describes Sentinel-2 as a multispectral optical mission, with bands at different spatial resolutions and applications including land management, agriculture, and disaster-related work. These characteristics shape what can be observed. They do not turn every image into a ground-level diagnosis. The measurement must be interpreted at the scale and under the conditions it supports.", "links": [ { "label": "Copernicus SentiWiki — Sentinel-2 mission", "href": "https://sentiwiki.copernicus.eu/web/s2-mission" } ] }, { "id": "section-3", "heading": "Connect the map to a field question", "body": "Consider a hypothetical team reviewing changes around a watershed. A useful workflow might identify areas for closer inspection, compare observations across dates, and combine them with local records. It should distinguish a signal worth investigating from a confirmed cause. Ask whether seasonal conditions, acquisition differences, or gaps in coverage could explain part of the apparent change before assigning a stronger interpretation." }, { "id": "section-4", "heading": "Design the handoff to the decision-maker", "body": "Our recommendation is to give each finding a location, time window, source, and statement of uncertainty. A map for prioritizing site visits needs different detail from a regional planning overview. Include a route back to the source observation and document important processing choices. If the user cannot tell which areas were observed and which were inferred, the graphic may communicate more confidence than the analysis supports." }, { "id": "section-5", "heading": "Keep visual polish proportional to evidence", "body": "Strong colors and precise boundaries can suggest certainty even when the underlying signal is ambiguous. Use legends, dates, and clear category definitions. Avoid presenting a conceptual illustration as a measured map. Geospatial work earns trust when it helps people ask better questions about a place—and makes clear what additional evidence is needed before taking action." }, { "heading": "The practical takeaway", "body": "Attach context and uncertainty to the map, not just a compelling color scale." }, { "heading": "Sources & further reading", "body": "Primary sources consulted for this analysis. Examples are hypothetical; recommendations are our editorial interpretation.", "links": [ { "label": "Copernicus SentiWiki — Sentinel-2 mission", "href": "https://sentiwiki.copernicus.eu/web/s2-mission" } ] } ] }, { "title": "A smaller architecture can leave more room to grow.", "url": "https://data-sorcerers.com/blog/modular-software-before-microservices/", "summary": "Splitting software into services is a deployment decision. Separating responsibilities is a design discipline that matters at every scale.", "sections": [ { "id": "section-1", "heading": "Complexity can arrive before the product does", "body": "A young product can acquire an elaborate service diagram before its team understands which features users need. Each separate service adds another boundary to operate, test, and reason about. That cost may be justified, but it should answer a concrete requirement. A diagram with many boxes is not, on its own, evidence of a system that will be easier to change." }, { "id": "section-2", "heading": "A useful architectural caution", "body": "In Monolith First, Martin Fowler discusses starting with a monolith and learning boundaries before moving toward microservices. He also emphasizes that decomposition is not automatically easy: internal modularity matters. We treat this as an architectural perspective, not a universal rule. Some systems have deployment or organizational constraints that justify separate services from the beginning.", "links": [ { "label": "Martin Fowler — Monolith First", "href": "https://martinfowler.com/bliki/MonolithFirst.html" } ] }, { "id": "section-3", "heading": "Make responsibilities explicit", "body": "Consider a hypothetical application combining enrollment, assessment, and certificates. A single deployment can still separate those responsibilities in its code, data access, and tests. Define what each module owns and what other modules may request from it. If every part directly modifies every other part’s tables, the system can become tightly coupled even when its files look neatly organized." }, { "id": "section-4", "heading": "Split for an observed reason", "body": "Our recommendation is to document the pressure that motivates a service boundary. It might be an independent release cadence, a distinct scaling need, or a security requirement. Compare that benefit with the work of coordinating versions, handling partial failures, and operating another deployment. A team should be able to explain which problem the separation solves and how it will know the change helped." }, { "id": "section-5", "heading": "Preserve options through ordinary engineering", "body": "Clear interfaces, focused modules, and meaningful tests can make future changes more manageable. So can documentation that explains why a boundary exists. Avoid designing every possible future service before the product provides evidence. A useful architecture allows the team to deliver today while retaining reasonable choices tomorrow. Growth should be a reason to revisit the design, not a requirement to make the first release unnecessarily complicated." }, { "heading": "The practical takeaway", "body": "Separate responsibilities early; separate deployments when there is a clear reason." }, { "heading": "Sources & further reading", "body": "Primary sources consulted for this analysis. Examples are hypothetical; recommendations are our editorial interpretation.", "links": [ { "label": "Martin Fowler — Monolith First", "href": "https://martinfowler.com/bliki/MonolithFirst.html" } ] } ] }, { "title": "An API is a promise to the team on the other side.", "url": "https://data-sorcerers.com/blog/api-contracts-for-change/", "summary": "Clear interfaces help systems evolve independently. The difficult work is deciding which promises must remain stable when the implementation changes.", "sections": [ { "id": "section-1", "heading": "The consumer sees behavior, not your refactor", "body": "A backend team can make a substantial internal improvement while accidentally breaking a client that depends on a small response detail. The client does not care that the database is cleaner or the implementation is faster if an expected field changes meaning. An API is therefore more than an endpoint list. It is an agreement about behavior between teams that may release on different schedules." }, { "id": "section-2", "heading": "Design around a stable interface", "body": "Microsoft’s API design guidance emphasizes platform independence, loose coupling, standard protocols, and resource-oriented interfaces. These principles help consumers work without depending on the provider’s internal implementation. They do not remove the need to make explicit choices about compatibility, errors, pagination, or the meaning of a response.", "links": [ { "label": "Microsoft Azure Architecture Center — Web API design best practices", "href": "https://learn.microsoft.com/en-us/azure/architecture/best-practices/api-design" } ] }, { "id": "section-3", "heading": "Describe the ordinary and exceptional cases", "body": "For a hypothetical enrollment API, define what happens when an application is received, when the same request is retried, and when a required field is missing. Explain which identifier the client should retain and how it can distinguish a temporary failure from an invalid request. Examples should cover incomplete and unsuccessful cases, not only the most attractive successful response." }, { "id": "section-4", "heading": "Make change visible to consumers", "body": "Our recommendation is to review changes from the consumer’s perspective. Adding an optional field is different from changing an existing field’s type or interpretation. Document deprecations and agree on a migration window where needed. Use tests that exercise real client expectations. A contract that exists only in a document can drift away from the behavior actually served in production." }, { "id": "section-5", "heading": "Keep ownership and support discoverable", "body": "A useful API reference explains authentication, limits, error conventions, and how to report a problem. It should identify who owns the interface and how changes are communicated. Consider the person integrating under time pressure: can they determine whether a request succeeded without guessing? Good API design reduces the amount of hidden organizational knowledge required to build on someone else’s work." }, { "heading": "The practical takeaway", "body": "Review an API change as a change to another team’s workflow, not only to your code." }, { "heading": "Sources & further reading", "body": "Primary sources consulted for this analysis. Examples are hypothetical; recommendations are our editorial interpretation.", "links": [ { "label": "Microsoft Azure Architecture Center — Web API design best practices", "href": "https://learn.microsoft.com/en-us/azure/architecture/best-practices/api-design" } ] } ] }, { "title": "Uptime is not the same as a working user journey.", "url": "https://data-sorcerers.com/blog/reliability-user-journeys/", "summary": "A service can answer requests while the task that matters to the user still fails. Reliability measures need to follow the experience.", "sections": [ { "id": "section-1", "heading": "The server is healthy. The application still feels broken.", "body": "A status page can be green while users cannot complete registration, retrieve a report, or submit a form. Infrastructure availability is important, but it is only part of the experience. A useful reliability discussion begins with the task the service exists to support. If the team cannot name that task precisely, it may be monitoring what is easiest to measure rather than what matters most." }, { "id": "section-2", "heading": "Connect indicators to an objective", "body": "Google’s Site Reliability Engineering book explains service-level indicators and objectives as a way to define and evaluate meaningful service behavior. Its discussion includes measures such as availability and latency, with attention to what users experience. The relevant lesson is to select an indicator that represents the service’s purpose and an objective that can guide operational decisions.", "links": [ { "label": "Google SRE Book — Service Level Objectives", "href": "https://sre.google/sre-book/service-level-objectives/" } ] }, { "id": "section-3", "heading": "Choose a journey you can observe", "body": "Consider a hypothetical application portal. Receiving an HTTP response does not prove that an application was durably saved or that the user received a usable reference. The journey may include upload, validation, persistence, and confirmation. Identify where failure can leave the user uncertain. Instrument the outcome without collecting more personal information than the operation requires." }, { "id": "section-4", "heading": "Agree on what happens when reliability falls", "body": "Our recommendation is to connect a reliability objective with an operating response. Who investigates a rise in failed submissions? When should a release pause? Which fallback keeps the user informed? Keep the objective realistic for the service and its dependencies. A target that is never used to make decisions can become a reporting ritual rather than a useful agreement." }, { "id": "section-5", "heading": "Learn from the path through an incident", "body": "After a failure, reconstruct what the user experienced and where the team first received a useful signal. Look for missing context, unclear ownership, and recovery steps that depended on one person. A stronger system may require a clearer error message or safer retry behavior as much as additional infrastructure. Reliability becomes meaningful when the user can finish the task and understand its result." }, { "heading": "The practical takeaway", "body": "Measure the completed user task, then connect the measure to a concrete operational response." }, { "heading": "Sources & further reading", "body": "Primary sources consulted for this analysis. Examples are hypothetical; recommendations are our editorial interpretation.", "links": [ { "label": "Google SRE Book — Service Level Objectives", "href": "https://sre.google/sre-book/service-level-objectives/" } ] } ] }, { "title": "When an AI assistant reads the web, it crosses a trust boundary.", "url": "https://data-sorcerers.com/blog/prompt-injection-trust-boundaries/", "summary": "Retrieved material can contain instructions as well as information. The system must keep outside content separate from authority to act.", "sections": [ { "id": "section-1", "heading": "Reading is not the same as receiving an instruction", "body": "An assistant that reads documents or web pages will encounter text written by people outside the application. Some of that text may describe a task; some may try to redirect the assistant. The system needs a clear distinction between information it can analyze and instructions it is authorized to follow. That distinction becomes especially important when the assistant can access tools or private data." }, { "id": "section-2", "heading": "The risk extends beyond a chat message", "body": "OWASP’s guidance describes direct and indirect prompt injection, including instructions embedded in external sources. It also notes that retrieval-augmented generation and fine-tuning do not fully mitigate this class of vulnerability. This means a trusted retrieval pipeline does not make every retrieved sentence trustworthy as an instruction.", "links": [ { "label": "OWASP Gen AI Security Project — LLM01:2025 Prompt Injection", "href": "https://genai.owasp.org/llmrisk/llm01-prompt-injection/" } ] }, { "id": "section-3", "heading": "Map what the assistant can reach", "body": "For a hypothetical research tool, list the data it can read and the actions its tools can perform. Searching public pages, reading internal documents, and sending an external message are distinct permissions. A document discovered during a search should not be able to expand those permissions. Keep authorization checks in application logic, where they can be inspected and tested, rather than depending entirely on a model’s interpretation." }, { "id": "section-4", "heading": "Test the boundary with representative cases", "body": "Our recommendation is to include benign task content and adversarial instructions in a controlled evaluation set. Check whether the assistant stays on task, protects restricted information, and declines actions outside its scope. Review tool calls as well as final text. A harmless-looking answer does not prove that an inappropriate intermediate action was prevented." }, { "id": "section-5", "heading": "Plan for containment, not a perfect prompt", "body": "Use limited tool permissions, explicit approval for consequential actions, and logs that support investigation without exposing unnecessary sensitive content. Give the operator a way to interrupt a task and revoke access. No single prompt should be presented as a complete security solution. The useful design question is what the system can still protect when a model interprets an untrusted instruction incorrectly." }, { "heading": "The practical takeaway", "body": "Keep authority in the application’s permission model, not in text retrieved from outside it." }, { "heading": "Sources & further reading", "body": "Primary sources consulted for this analysis. Examples are hypothetical; recommendations are our editorial interpretation.", "links": [ { "label": "OWASP Gen AI Security Project — LLM01:2025 Prompt Injection", "href": "https://genai.owasp.org/llmrisk/llm01-prompt-injection/" } ] } ] }, { "title": "A fast interface is only useful if people can operate it.", "url": "https://data-sorcerers.com/blog/performance-accessibility-shared-design/", "summary": "Performance and accessibility meet in the same user journey: finding information, understanding controls, and completing a task without unnecessary friction.", "sections": [ { "id": "section-1", "heading": "The score is not the whole experience", "body": "A web page can load quickly and still be difficult to navigate with a keyboard. It can have carefully labeled controls and still freeze during an important action. These are different engineering problems, but they meet in the experience of the same person. Reviewing them separately should not mean losing sight of the complete task." }, { "id": "section-2", "heading": "Use complementary measures", "body": "Google’s Core Web Vitals guidance focuses on loading, responsiveness, and visual stability. WCAG provides accessibility criteria and techniques for making content perceivable, operable, understandable, and robust. These frameworks answer different questions. A high performance score is not evidence of accessibility conformance, and an automated accessibility scan does not establish that every user journey works.", "links": [ { "label": "Google web.dev — Web Vitals", "href": "https://web.dev/articles/vitals" }, { "label": "W3C — How to Meet WCAG 2.2", "href": "https://www.w3.org/WAI/WCAG22/quickref/" } ] }, { "id": "section-3", "heading": "Follow one task across different conditions", "body": "Consider a hypothetical mobile application form. Test whether labels remain visible while typing, errors explain how to recover, and focus reaches the relevant control. Then examine how the page behaves on a slower device or connection. An upload button that shifts position or a success state that appears late can create uncertainty even when the form is technically functional." }, { "id": "section-4", "heading": "Choose improvements that remove friction", "body": "Our recommendation is to prioritize changes that help the actual task: smaller images, stable layout dimensions, clear headings, semantic controls, and manageable JavaScript work. Preserve useful HTML before client-side behavior becomes available. Avoid postponing essential interaction merely to improve a benchmark. A performance improvement should make the page easier to use, not make the test easier to satisfy." }, { "id": "section-5", "heading": "Combine laboratory checks with human review", "body": "Automated audits are useful for repeatable checks and regressions. Add keyboard navigation, zoom, screen-reader review where available, and tests on realistic devices. Record which routes and states were examined rather than reporting one score as a guarantee for the entire website. The release question is whether people can understand and complete the work—not whether a dashboard has turned entirely green." }, { "heading": "The practical takeaway", "body": "Test a complete journey under real constraints, using performance and accessibility measures together." }, { "heading": "Sources & further reading", "body": "Primary sources consulted for this analysis. Examples are hypothetical; recommendations are our editorial interpretation.", "links": [ { "label": "Google web.dev — Web Vitals", "href": "https://web.dev/articles/vitals" }, { "label": "W3C — How to Meet WCAG 2.2", "href": "https://www.w3.org/WAI/WCAG22/quickref/" } ] } ] }, { "title": "Responsible AI becomes real in ordinary product decisions.", "url": "https://data-sorcerers.com/blog/responsible-ai-operating-decisions/", "summary": "Governance matters when it changes what a team builds, how it evaluates the result, and who can intervene when the system causes problems.", "sections": [ { "id": "section-1", "heading": "A policy cannot make every product decision", "body": "An organization may publish thoughtful AI principles and still leave a delivery team uncertain about a specific feature. May the system use this information? Who reviews an error? Is an automated recommendation appropriate for this task? Governance becomes useful when people can answer these questions while building and operating the product, not only during a presentation about values." }, { "id": "section-2", "heading": "A framework for organizing the conversation", "body": "NIST’s AI Risk Management Framework provides a voluntary approach to managing AI-related risks. Its core functions—Govern, Map, Measure, and Manage—organize work across context, assessment, and response. The framework is not a certification that a product is safe. It can help a team structure questions and responsibilities that must be resolved in its own setting.", "links": [ { "label": "NIST — AI Risk Management Framework", "href": "https://www.nist.gov/itl/ai-risk-management-framework" } ] }, { "id": "section-3", "heading": "Start with the people affected", "body": "For a hypothetical learning assistant, the immediate user may be a teacher, but students may be affected by its output. Describe the task, the data involved, and the consequences of an incorrect recommendation. Identify who can challenge or correct the result. Different applications require different review arrangements; a convenience feature and a consequential decision should not inherit the same operating assumptions by default." }, { "id": "section-4", "heading": "Turn principles into release evidence", "body": "Our recommendation is to connect each important concern with a check and an owner. If the product promises source-based answers, test whether the sources support them. If users need human review, make that route visible and workable. Document limitations in language the operator can use. Keep a record of unresolved issues so a release decision reflects the actual system rather than an idealized description." }, { "id": "section-5", "heading": "Keep the review alive after launch", "body": "Models, data, users, and surrounding processes can change. Review incidents and feedback as signals about the operating design, not merely as isolated mistakes. Give the team authority to pause or narrow a feature when evidence warrants it. Responsible AI is sustained through these ordinary choices: what information is collected, which actions are permitted, how failures are handled, and whether people retain meaningful oversight." }, { "heading": "The practical takeaway", "body": "Make each important AI principle visible in a product decision, an evaluation, and an accountable owner." }, { "heading": "Sources & further reading", "body": "Primary sources consulted for this analysis. Examples are hypothetical; recommendations are our editorial interpretation.", "links": [ { "label": "NIST — AI Risk Management Framework", "href": "https://www.nist.gov/itl/ai-risk-management-framework" } ] } ] } ], "platforms": { "careers": "https://careers.data-sorcerers.com", "pyronyx-academy": "https://academy.data-sorcerers.com", "pyronyx-certification": "https://certification.data-sorcerers.com", "llm-integration": "https://arutala.data-sorcerers.com", "sarl-sorcery-ai-racing-league": "https://sarl.data-sorcerers.com", "data-sorcerers-community": "https://community.data-sorcerers.com", "herai": "https://her-ai.data-sorcerers.com", "her-ai": "https://her-ai.data-sorcerers.com" } }