rag-optimization

RAG optimization auto-adjusts retrieval, ranking, embedding, and vector-search stages to enhance relevance and speed. Configure hybrid weights (vector, keyword, graph), rerank thresholds, top-K results, query expansion counts, and semantic caching to balance precision against latency across document QA and knowledge-base search.

Hybrid weighting combines vector, keyword, and graph search by assigning configurable weights to each. This approach captures semantic similarity, exact matches, and relational context simultaneously, improving relevance when no single retrieval method alone covers all query intents.

Query expansion generates multiple reformulations of the original query to capture varied phrasings and intents. Re-ranking uses Cross-Encoder thresholds to filter and order results by confidence, removing low-relevance matches and surfacing the most precise answers from retrieved documents.

Semantic caching stores results for similar queries and reuses them when cache hit rates are high, reducing computation. The Skill monitors cache performance and supports invalidation to ensure stale or changed document content doesn't return outdated answers.

Yes. The Skill applies to document QA, knowledge-base search, and multi-source information retrieval by tuning shared retrieval, ranking, embedding, and vector-search stages—the common infrastructure across these scenarios.

Tune vector weight, keyword weight, graph weight, rerank threshold, top-K result count, and query expansion count. Start with vector_weight=0.6, keyword_weight=0.4, then test expanded queries and re-ranking thresholds against your corpus to balance precision and recall.