Get the latest tech news
Better RAG Results with Reciprocal Rank Fusion and Hybrid Search
The problem with vector-only search At Assembled, our issue resolution engine is designed to assist customer support by suggesting potential answers to support queries. We use Retrieval Augmented Generation (RAG) for much of this pipeline because it's quicker to iterate on than fine-tuning, doesn’t require training on customer data (which many companies prefer), and generally provides high-quality results.
We use Retrieval Augmented Generation (RAG) for much of this pipeline because it's quicker to iterate on than fine-tuning, doesn’t require training on customer data (which many companies prefer), and generally provides high-quality results. Customer support teams often have multiple articles on similar topics and lack a tightly curated knowledge base, leading vector search to sometimes return irrelevant results to our RAG engine and reduce response accuracy. Since implementing this framework, there have been a lot of developments in RAG-based techniques, such as fine tuning of embedding models, applying matrix transformations on vector results, HyDE, etc.
Or read this on Hacker News