Inspiration
The question that started everything was simple but unsettling: what happens to the accountant, the lawyer, the developer, the data scientist - when AI can do the core of their job?
The answer most people give is "they get replaced." We refused to accept that.
We watched graduates leave university with strong academic knowledge but zero fluency in the tools that employers were already deploying. We watched companies adopt AI at record speed - AI adoption in accounting firms alone quadrupled from 9% to 41% in a single year - while their junior hires had no idea how to work alongside those tools. We watched the skills gap widen in real time, across every profession, not just in tech.
And then we asked a different question: what if the rise of AI could create more human opportunity, not less?
That question became AI Craft.
The insight at the heart of it is this - AI is not destroying work. It is destroying the old way work was done. The professionals who become irrelevant are not the ones AI replaces. They are the ones who refused to adapt. The accountant who learns to govern AI output, catch what it misses, and advise clients on what the numbers mean in context becomes more valuable, not less. The developer who learns to direct Kiro, design the architecture, and integrate the systems that AI cannot wire together on its own becomes a senior engineer from day one.
AI Craft exists to make that shift accessible to every student, graduate, and young professional - regardless of where they are, what they studied, or which profession they are entering.
What it does
AI Craft is an AI-native skills platform that trains professionals not to compete with AI, but to lead it.
A student opens AI Craft and uploads their academic documents - their syllabus, textbook chapters, lecture notes. Our platform indexes that content and generates role-specific interview questions calibrated exactly to what that student studied, not generic content that misses their curriculum entirely.
Then the student enters a real industry task. Not a tutorial. Not a video. A task that mirrors what their first employer will actually ask them to do on day one. The AI tools - Kiro for developers, Copilot for Finance for accountants, Julius AI for data scientists - are already active inside the task environment, handling what AI handles: code generation, data analysis, document processing.
The student's job is what AI cannot do. Designing the database schema. Deciding which model to integrate and why. Interpreting the financial anomaly the AI flagged. Making the architectural decision. Exercising the professional judgment that no tool can replicate.
Every task has two evaluation layers. The first is technical - did the student complete the deliverable correctly? The second is human - did the student make sound decisions, demonstrate professional judgment, and understand why the AI did what it did?
This is the dual-layer curriculum that makes AI Craft different from every other platform. We do not teach students what AI will do for them. We teach them what they must become in order to lead it.
AI Craft covers every knowledge profession:
- Full stack developers - Kiro IDE, system architecture, database integration, security judgment
- Data scientists - model governance, bias auditing, deployment decisions, business interpretation
- Accountants - Copilot for Finance, professional skepticism, audit oversight, client advisory
- Legal professionals - Harvey AI, CoCounsel, ethical reasoning, advocacy judgment
- Marketers - AI-assisted strategy, brand direction, creative oversight
- UX designers - Figma AI, user research, accessibility, stakeholder negotiation
- Healthcare workers - clinical AI tools, diagnosis nuance, patient-centered judgment
- HR professionals - AI-assisted hiring, culture sensing, conflict resolution
How we built it
The current prototype is built on a stack that proved the core concept works.
We used Anthropic Claude as the reasoning engine - the model that reads a student's academic document, understands its structure, and generates interview questions and industry tasks that are genuinely calibrated to that specific content. Not generic prompts. Contextually aware, curriculum-specific output.
Inngest handles our backend workflow orchestration - managing the asynchronous pipeline from document upload through content extraction, question generation, and task delivery.
The document ingestion pipeline uses RAG (Retrieval-Augmented Generation) principles - the student's academic content becomes the knowledge base from which all questions and tasks are generated, ensuring that what the platform produces is always grounded in what the student actually studied.
The frontend is built to be simple and task-focused -- upload a document, receive questions, receive a task, complete it, get evaluated.
What we built is a working proof that the core pipeline functions: document in, relevant professional output out. The prototype demonstrates the concept. The next phase - integrating Microsoft Foundry Agent Service for multi-agent orchestration, Foundry IQ for production-grade RAG, and Kiro IDE for real embedded tool environments - will transform that proof into a full platform.
Challenges we ran into
Making evaluation meaningful, not just mechanical. Generating interview questions is straightforward. Generating good interview questions - ones that distinguish a student who truly understands a concept from one who has memorised a definition - required significant prompt engineering and iteration. Generic questions are easy. Questions that test the judgment layer are hard.
The gap between academic content and industry reality. A student's syllabus talks about concepts. An industry task requires applying those concepts in messy, real-world conditions. Bridging that gap - taking academic content and generating tasks that feel genuinely professional rather than like glorified homework - was harder than we expected and remains an area of active improvement.
Scope clarity. The temptation when building something this broad is to build everything at once. We learned early that a platform for eight professions with shallow coverage is worth far less than a platform for two professions with deep, genuine value. Discipline about scope - starting with the professions we know deeply, full stack development and data science - was a challenge and ultimately the right decision.
Building for a mission that is bigger than a feature. AI Craft is not trying to be a better Udemy. It is trying to change the relationship between an entire generation and the AI systems that are reshaping their working lives. Keeping that mission central while making pragmatic, incremental product decisions is an ongoing tension that we navigate every day.
Accomplishments that we're proud of
We built a working prototype that demonstrates the most important thing we needed to prove: a student's own academic content can be transformed into genuinely relevant professional preparation.
That sounds simple. It is not. Most platforms offer generic content and call it personalised. AI Craft's document-aware pipeline generates questions and tasks that are specifically grounded in what that individual student actually studied - not a one-size-fits-all question bank.
We are proud that the prototype works across disciplines. A computer science student and a data science student upload different documents and receive meaningfully different, domain-appropriate output. The architecture generalises.
We are proud of the clarity of the idea. The dual-layer curriculum - AI tool fluency plus human judgment - is a framework that every employer we have spoken to immediately recognises as exactly what they need from their junior hires. That clarity means the platform sells itself when you describe it.
And we are proud of the mission. We are building this from Pretoria, South Africa, for a continent that will produce more new workforce entrants than the rest of the world combined by 2050. If AICraft works, it does not just help students get jobs. It helps an entire generation leapfrog the legacy tooling debt that slows workers in the developed world and enter the AI-native global workforce directly.
What we learned
The problem is more urgent than we thought. When we started, we believed the AI skills gap was a future problem. It is not. It is happening now. AI adoption quadrupled in accounting in a single year. Legal AI is already deployed in major firms. Companies are not waiting for a generation of AI-fluent graduates - they are hiring the graduates they have and watching them struggle to adapt. AI Craft is not early. It is needed today.
The human judgment layer is the product. We initially thought the AI tool training was the core offering. We learned that the harder, more valuable, and more defensible part is teaching the irreplaceable human skills that sit alongside those tools. Any platform can teach someone to use Kiro. Only a platform that deeply understands each profession can teach a developer when to override Kiro's architectural decision, or teach an accountant when to challenge what the AI flagged as clean.
Africa is not a developing market for this idea - it is the leading market. African graduates do not have decades of legacy tooling and legacy mindsets to unlearn. They can enter the AI-native global workforce without the baggage that slows adaptation in the developed world. That is not a disadvantage. That is a structural advantage that makes Africa the ideal place to build and prove this platform before scaling globally.
Starting narrow is not thinking small. Covering two professions deeply is more valuable than covering eight shallowly. We learned that the quality of a single task - the specificity of its brief, the realism of its context, the rigour of its evaluation rubric - determines whether a student feels genuinely prepared or just entertained. Depth wins.
What's next for AI Craft
The prototype proves the concept. The next phase builds the platform.
Immediate (0–3 months): We migrate from Inngest to Microsoft Foundry Agent Service for production-grade multi-agent orchestration. We integrate Foundry IQ for a proper RAG pipeline over student academic documents. We integrate Kiro IDE - with full terminal and CLI - as the embedded development environment for the full stack developer track, replacing the current placeholder task environment. The document-to-interview-question pipeline gets upgraded from prototype quality to production quality.
Short term (3–6 months): We launch a 50-student beta with final-year students from South African universities. We measure task completion rates, pre/post confidence scores, and employer-readiness assessments. We iterate on the task quality based on real student feedback. We release the full interview prep module as a standalone feature - the strongest acquisition tool we have.
Medium term (6–12 months): We launch the Data Science track with 25 real industry tasks, model deployment to Foundry endpoints, and a governance module. We launch individual subscriptions and the employer talent portal, where anonymised graduate profiles are browsable by companies seeking AI-fluent junior hires. We close the first institutional pilot with a South African university or bootcamp.
Long term (12–18 months): We add the Accounting, Legal, and UX profession tracks with industry advisors from each field. We begin pan-African expansion to Nigeria, Kenya, and Ghana. We raise a Series A backed by measurable traction: student completions, employer placements, and institutional contracts.
The mission does not change at any stage: ensure that the rise of AI creates more human opportunity, not less. Teach every professional not what AI will do for them, but what they must become in order to lead it.
AI Craft is four months from an MVP. It is eighteen months from pan-African scale. And it is one generation away from being the platform that proved Africa could leapfrog the world into the AI-native era.
We are building it from Pretoria. We are building it now.
Log in or sign up for Devpost to join the conversation.