AI learning
Build the model. Test the guardrails.
Learn how language models train, how retrieval grounds an answer, and how to keep AI-assisted changes within a tested boundary. Start with a free project and leave with working files you can inspect, change and keep.
Two free projects. Real work to take away.
No login, installation or API key required.
Make your own LLM
Build a tiny neural next-token model in JavaScript. Follow the numbers from tokens to predictions, gradients and changed weights.
- Tokenize a corpus and learn embeddings.
- Check your gradients and measure loss on held-out text.
- Generate text, save your trained weights and restore the model.
You keepYour browser app, trained checkpoint, tokenizer and experiment notes.
Bring JavaScript functions, arrays and basic algebra. This is a small teaching model; it does not produce a general-purpose chatbot.
Train your model
Agentic guardrails
Turn a plausible AI code proposal into a change you can trust enough to review. Set the scope, find the defect and prove the repair preserves valid behavior.
- Write an issue, repository instructions and permission rules.
- Catch a bad proposal with your own regression test.
- Review changed files, stale evidence and an unfamiliar handoff.
You keepA repository ZIP with your app, tests, CLAUDE.md, settings and review notes.
Bring basic JavaScript. Your exported folder is prepared for Claude Code; that optional tool is separate from this browser project.
Build your guardrailsGo deeper with the full courses
Both subjects are included in Web, Plus and Pro.
LLM & RAG engineering
Connect model foundations and training to retrieval, grounded answers and evaluation. Practice access control and failure analysis so an answer is supported by evidence the user is allowed to see.
Model foundations · Training · Retrieval · Evaluation · Access control
Agentic programming
Design tasks and context for AI-assisted work, control tools and permissions, and verify what actually happened. Trace bounded execution and review completion against the requested outcome.
Task design · Context · Tools · Permissions · Verification · Traces
Theory & reasoning — read explanations, predict behavior and examine failure cases before committing to a design.
Hands-on coding — implement a change, run checks, inspect the evidence and tackle a changed problem.
Want an assistant to guide your session? Find MCP learning on Agents.
