Ranks and filters retrieved documents based on vector similarity metrics to optimize RAG pipeline relevance.
The Document Ranking skill provides a specialized utility for sorting and prioritizing search results within an Adaptive RAG (Retrieval-Augmented Generation) workflow. By processing distance and similarity metrics already computed during vector search, it enables precise control over document relevance without redundant database queries. This skill is essential for developers looking to improve the quality of LLM responses by ensuring only the most contextually relevant information is passed to the generation phase, effectively reducing token costs and noise.
Key Features
01Sorts documents by similarity score or cosine distance
02Filters retrieval results using customizable similarity thresholds
03Extracts top-k documents to minimize context window usage
04Supports collection-priority ranking for multi-source retrieval
05Seamless integration with grading and generation workflows
060 GitHub stars
Use Cases
01Optimizing RAG context relevance for high-accuracy AI responses
02Prioritizing specific documentation sources in multi-collection searches
03Reducing LLM API costs by filtering out low-confidence search results
