| Beschreibung |
Retrieval augmented generation (RAG) combines (web) search and generative AI to provide fast and reliable information access. In this project, we want to implement such a RAG component for our own search engine, www.chatnoir.eu. However, RAG systems are very expensive as each query triggers a web search and text generation with a long prompt. To avoid long and expensive searches, we want to develop a multi-stage caching system that, firstly, stores and retrieves answers given to near-duplicate queries, and secondly, caches partial answers and composes responses to new queries from those. |
