Understand AI
by seeing it work.
Interactive lessons for AI engineering. Explore token generation, context windows, RAG, agents, and release evaluation through visual simulations.
Start with three interactive foundations
Change a parameter, watch the system respond, and connect the result back to an engineering decision.
Your local progress
Stored in this browser. No account needed.
Token Playground
See how LLMs predict the next token.
Context Window Lab
See how context fills, overflows, and gets managed.
Agent Loop Simulator
Step through act → observe → refine.
Production Lab previews
Start from a failure state. Tune the engineering policy, compare against the baseline, and learn what the system trades away when one metric improves.
Instruction Conflict Lab
Fix instruction authority, then see why Prompt still cannot enforce runtime safety.
RAG Failure Lab
Fix retrieval without assuming more context is always better.
Context Compression Lab
Save tokens without silently deleting the information that changes the answer.
Agent Reliability Lab
Improve reliability without pretending retries, guardrails, and approvals are free.
Evaluation Failure Lab
Find the release regression hidden behind a better aggregate score.
See. Play. Break. Aha. Build.
See
Make invisible AI states visible through clear diagrams and simulations.
Play
Change parameters and compare outcomes instead of memorizing vocabulary.
Break
Trigger failure modes intentionally so the weak assumptions become visible.
Aha
Connect cause and effect until the system clicks as a durable mental model.
Build
Apply the mental model to architecture, trade-offs, evaluation, and production decisions.
Six responsibilities behind production AI systems
The layers are engineering responsibilities, not framework names. Learn where a failure belongs before deciding how to fix it.
Go from AI curiosity to AI engineering.
Join early access for new failure simulations and production labs.