Build vector search and embeddings

Optimize search, indexing, and vector database retrieval

Measures recall and precision before/after every change—catches regressions that impact search quality invisibly.

• · existing retrieval code
• · queries and test data

• · hybrid search implementation
• · reranker config
• · query expansion rules
• · performance baselines

• · Qdrant, Pinecone, or Weaviate
• · embedding models
• · reranker models

Task owned_files specified; recall/precision baselines known

Low recall if chunking too aggressive; latency regression if index too large; precision collapse with bad reranker

• · tests for query construction and filtering
• · latency profiling for each path
• · configuration in config files not hardcoded