AI School 369 logoAI School 369Production AI, explained simply

Projects

The school's rule: if we teach it, we ship the code. Every project is public, MIT licensed, small enough to read in one sitting, and verified by a real run before it is linked here.

projects/rag-freshnessCompleteRAG Freshness: Catch a Stale Index LyingA 260-line demo that builds an index, edits a document behind its back, and watches the chatbot answer from the old version. Then the fix: version stamps, hash-based refresh, deletion propagation, and 20 canary questions.Receipt: Verified run: 20/20 canaries pass on the fresh index, the stale index answers '30 days' after the policy changed to 60, refresh fixes it, deleted documents stop answering.projects/agent-tool-loopCompleteAgent Tool Loop: Budgets, Retries, a Circuit BreakerThe think-act-observe loop in about 220 lines of standard-library Python, with the three controls production loops need: a step budget, a per-tool retry budget, and a circuit breaker.Receipt: Verified run: healthy task completes in 4 calls with a flaky tool succeeding on attempt 3; the stuck task's breaker trips at step 3 of 12.projects/reranking-labIn the labReranking LabTwo-stage retrieval measured honestly: the same questions, the same corpus, with and without the reranking pass.Receipt: Ships with its newsletter edition; the repo README tracks progress.projects/agent-memory-sqlIn the labAgent Memory in SQLLong-term agent memory you can inspect with SELECT statements, instead of a vector store you have to take on faith.Receipt: Ships with its newsletter edition; the repo README tracks progress.projects/canary-evalsIn the labCanary EvalsA tiny eval set that runs after every change and fails loudly, so quality regressions arrive as alarms, not user reports.Receipt: Ships with its newsletter edition; the repo README tracks progress.projects/confirmation-gatesIn the labConfirmation GatesBlast-radius design for agents: which actions run free, which need a human yes, and how the gate is enforced in code.Receipt: Ships with its newsletter edition; the repo README tracks progress.

The whole library: github.com/urbanschool369-oss/ai-school-369-projects. Run them, break them, open an issue when you find something we missed.