Project: Top-K Retrieval Policy Planner
Build a planner that chooses retrieval policy from query intent, risk, score distribution, and trace signals.
What you will build
- A query-intent to top-k policy mapper.
- A retrieval miss diagnostic engine.
- A controls recommender for filters, rewrite, multi-query, dedupe, and reranking.
- A trace report explaining the retrieval decision.
Product-level outcome
Evidence selection policy
Your project should output a practical retrieval policy: "For high-risk policy exceptions use k=10, EU active filters, query rewrite, parent expansion, dedupe, and reranking."