← back
The State of AI Powered Search and Retrieval — Frank Liu, MongoDB (prev Voyage AI)
Takeaway
Production AI search now means domain-specific embeddings, structured-filter hybrid retrieval, and agentic query decomposition — not one-shot vector search.
Summary
- Liu (Voyage AI, now MongoDB) defines AI-powered search as understanding intent and related concepts beyond BM25, including reasoning and instruction-following.
- Three real-world lessons: no one-size-fits-all embedding model (continue.dev evaluated and picked Voyage-Code-3 for codebase chat); embeddings alone aren't enough — structured metadata filters (state, doc type) matter; agentic retrieval is a feedback loop (LLM decomposes 'Q4 2024 earnings' into Q1/Q2/Q3/Q4 sub-queries).
- 2025/26 is the era of agents — embeddings need to be strong on conversational data, and retrieval systems need iterative query rewriting/decomposition rather than one-shot input/output.
- Voyage offers domain-specific models (Voyage-Code, Voyage-Law) plus rerankers, available via API, AWS, Azure, now part of MongoDB.
embeddingsvoyage-aiagentic-retrieval
Original description
In this talk, we examine the state-of-the-art in AI-powered search and retrieval. We detail techniques for enhancing performance beyond base embedding models, including hybrid search, reranking strategies, query decomposition and document enrichment, the use of domain-specific and fine-tuned embeddings, custom data processing pipelines (ETL), and contextualized chunking methods. About Frank Liu Frank Liu is Staff Product Manager at MongoDB. He has over a decade of industry experience in machine learning and hardware engineering and presents at major industry events like the Open Source Summit Open Data Science Conference. In his spare time, he enjoys experimenting with and training models. Frank holds MS and BS degrees in Electrical Engineering from Stanford University. Recorded at the AI Engineer World's Fair in San Francisco. Stay up to date on our upcoming events and content by joining our newsletter here: https://www.ai.engineer/newsletter