One final build that brings the whole course together: a complete evaluation and observability harness for Docent — golden dataset, deterministic and judge-based metrics, RAG diagnosis, tracing and cost accounting, and a production gate, all in one runnable pipeline.
Welcome to the Capstone, the final stop in the course. Every module built one instrument for understanding Docent. This project puts them on the same bench: a single evaluation-and-observability harness that takes Docent from a raw assistant to one you can measure, explain, trace, and gate — all in one runnable pipeline.
You’ll assemble the versioned golden dataset from Module 2, score answers with the deterministic metrics of Module 3 and the calibrated judge of Module 4, run the RAG diagnosis from Module 5 to separate retrieval failures from generation failures, wrap calls in the tracing and cost accounting of Module 7, and finish with the production-readiness gate and guardrails of Module 8. A final report traces every number back to the module that taught it — the whole course, expressed as one working system.
Start the lesson to build the harness that ties everything together.
Complete all 1 lessons to finish the Capstone module.