The State of AI Powered Search and Retrieval
summary
Overview of practical results and engineering choices for AI search: RAG plus reranking is standard, embeddings drive relevance and embedding quality is critical — presenter claims 95–99% of systems use embeddings; code search users (continue.dev) pick Voyage Code 3 after evaluation. Systems combine vector filters with structured metadata, and agents decompose queries (example: splitting “2024 earnings” into Q1–Q4 queries) for multiple vector stores. Roadmap items include multimodal embeddings (Voyage Multimodal 3), instruction-tuned vectors, and an agent-native database combining embedding, rerank and ETL.
AI-written summary · about
chapters (10)
- 0:00Introduction and agenda
- 1:23What is AI powered search
- 3:50RAG and embedding quality
- 4:55Chatting with codebases
- 5:56Structured data in search
- 7:21Agentic retrieval loops
- 9:08Future of multimodal search
- 10:31Instruction tuning in embeddings
- 11:33Agent native databases
- 12:14Q&A and closing
related talks
48mEnter the Brave New World of GenAI with Vector Search
47mBoosting AI With OpenSearch: Vector Stores, Similarity Search, and Retrieval-Augme...
06mWhat makes vector databases essential for RAG and AI apps?
27mMachine Learning for Advanced AI Search
30mBeyond Keywords: Building an AI-Powered Search That Reads Your Mind
39mAI-Powered Search in Modern E-Commerce
40mAgentic Retrieval: Building Self-Optimizing Search Systems
29mThe Art of Production Quality Retrieval