agency-data-vector-store-manager

Improving RAG retrieval quality involves tuning the vector index, ensuring embeddings are consistent and up to date, and evaluating whether retrieved documents are relevant to queries. This Skill focuses on exactly those three areas: index optimization, retrieval quality, and embeddings management.

Vector index optimization means adjusting index structure and parameters so similarity search stays accurate and performant as the corpus grows. The Skill guides reviewing the current index configuration and applying optimizations appropriate to the stored embeddings.

Irrelevant RAG results usually come from stale or inconsistent embeddings, a poorly tuned index, or a mismatch between query and document embedding strategies. Auditing embedding freshness and index configuration typically identifies the cause.

Embeddings should be regenerated when the embedding model changes, when source documents are updated, or when retrieval quality degrades over time. Keeping embeddings consistent across the corpus is essential for reliable similarity search.

Vector retrieval can miss exact-keyword matches and struggles with very domain-specific terminology if embeddings were not trained on similar data. In those cases, hybrid approaches combining keyword search with vector similarity may be more appropriate.