Boosting RAG with Hybrid Search: Keyword + Vector for Maximum Relevancy

Session Track: AI Engineering

Session Time: 02:00 pm - 02:30 pm November 6

Session description

As RAG systems become foundational to LLM and GenAI applications, the quality of the retrieval layer is more critical than ever. Traditional keyword-based search often struggles with ambiguous queries or unseen vocabulary, while pure vector search may return irrelevant or overly general results. In this session, we explore how hybrid search — combining lexical methods like BM25 with vector approaches such as k-NN and HNSW — dramatically improves the accuracy and relevancy of RAG outputs. We’ll highlight real-world applications across industries, from enhancing content recommendations to improving fraud detection and game content analysis. The session will also cover integration strategies, technical trade-offs, and performance considerations when deploying hybrid retrieval at scale. Whether you're building production LLM systems or exploring Gen AI capabilities, you’ll gain practical insights into unlocking more relevant, reliable results with hybrid search and retrieval.