Vector Databases and the Data Structure of AI ft. MongoDB’s Sahir Azam

MongoDB’s Sahir Azam argues vector databases are becoming the memory and state layer for enterprise AI, not a replacement for traditional databases.

• MongoDB built vector search after enterprise customers refused to run separate databases for full-text and semantic search side-by-side.
• A European automaker cut car-diagnosis time from hours to seconds using audio embeddings matched against a corpus of known fault sounds.
• Novo Nordisk now drafts clinical study reports (CSRs) in minutes with an LLM; previously took weeks of manual effort.
• Azam argues vectors are a settled primitive like B-tree indexes; the unsolved problem is hitting 99.99% retrieval quality in probabilistic systems.
• Vector, graph, and relational modalities are complements, not substitutes — enterprise RAG quality requires all three in one system.
• MongoDB’s Atlas cloud transformation required incentive spiffs, per-deal sales-team shadowing, and treating it as a full business reinvention, not a new SKU.
• Azam believes every application developer will own AI integration, displacing the old centralized ML-team model.

2025-02-13 · Watch on YouTube