Learn AI the way
you'd build it
No ML background needed. One idea runs the whole field — predict the next chunk of text, then repeat — and everything else is a refinement of it. This course is T-shaped: skim the whole map shallow first, then drive each topic deep through Bloom's six levels, ending where you build something real.
Start here. Broad and shallow — build intuition and a map of the whole field.
Play with the one idea that explains all of AI: next-token prediction. Tokens, temperature, embeddings, attention, RAG — all interactive.
THE FOUNDATION →A map of the entire AI-engineer field: nine domains, where you're strong vs. where to focus, and the Bloom ladder for going deep.
THE MAP →Get your bearings — the tools, the timeline, and the vocabulary drill.
The mid-2026 AI stack, layer by layer, with a recommended pick at each level.
TOOLS →A week-by-week path from where you are to interview-ready.
TIMELINE →Flip-card drill for the core vocabulary. Run it daily until it's automatic.
DAILY REPS →When to reach for it — and when prompting, RAG, or structured outputs get you there cheaper.
ORIENT →Each topic driven through Bloom's 6 levels: Remember → Understand → Apply → Analyze → Evaluate → Create.
The three that compound into a flagship: a RAG app, an eval suite on it, and an agent on top.
Make it real: fast, cheap, remembering, and hard to attack.
Where time and money go; the levers that turn a demo into something shippable.
L1→L6 →Agents that don't forget — memory as RAG over their own past.
L1→L6 →Trust nothing the model reads: prompt injection, least privilege, red-teaming.
L1→L6 →Cheapest model that works: gateways, fallbacks, cost-aware routing.
L1→L6 →Built T-shaped: the wire first, then deeper over time. Concepts are durable; frameworks are swappable — so learn the ideas deeply and treat tools as interchangeable.
Open-source · deploy your own on Vercel.