Exercises: Build a Pipeline Trace
Use starter.py to model offline and online RAG pipeline stages, then produce a trace that explains what happened.
1 Define offline stages
Represent parse, clean, chunk, metadata enrichment, indexing, validation, and publish.
2 Define online stages
Represent routing, permission filtering, retrieval, reranking, context assembly, answer policy, and tracing.
3 Add gates
Flag weak retrieval, missing required source family, denied access, stale index, and high-risk escalation.
4 Produce a production trace
The trace should include route, filters, source ids, answer status, first failed gate, and recommended next action.