What problem does it solve? Dense retrievers often miss relevant documents that fall outside the initial ANN search pool, and reranking a large candidate pool is expensive. This Skill guides you through using Qdrant's Relevance Feedback API to distill a feedback model's knowledge into the vector search itself, surfacing relevant documents the initial retrieval missed. ## Core Features & Use Cases - Weight Calibration: Train the RF formula weights (a, b, c) on 50-200 representative queries using the qdrant-relevance-feedback Python library before inference. - High-Precision Top Results: Score only 5 seed documents to achieve reranking-level top-1/top-3 precision at roughly one-fifth the scoring cost. - Recall Beyond the Initial Pool: Run a two-pass feedback workflow to discover relevant documents completely outside the initial ANN retrieval set, ideal for legal, medical, and compliance research. - Use Case: Your semantic search returns mediocre results and reranking 25 documents per query is too costly. Use this Skill to calibrate RF weights, then score only 5 seeds per query while matching reranking quality at the top of the ranking. ## Quick Start Help me set up Qdrant's Relevance Feedback API to improve my dense vector search relevance as a cheaper alternative to reranking.
