Writing

Introducing Lexical Prefilters for Vector Search, combining fuzzy text, phrase, wildcard, and geo filters with vector similarity search to improve precision, performance, and cost at query time.

Comprehensive benchmarks for Atlas Vector Search across datasets exceeding 10M vectors, showing how quantization can maintain 90–95% accuracy with sub-50ms latency at the highest dimensionalities.

A new Python library that unifies document processing, multimodal embedding generation, and vector indexing to enable search over text, images, charts, and PDFs with a single interface.

Atlas now combines full-text and vector search natively via $rankFusion and $scoreFusion aggregation stages, eliminating the need for separate search engines while improving relevance for RAG and AI agent use cases.

Atlas Vector Search introduces vector quantization to compress embedding sizes and reduce storage costs while preserving semantic fidelity, making large-scale AI search viable at lower cost.

Exact nearest neighbor (ENN) search guarantees retrieval of the absolute closest vectors, ideal for small datasets, ANN accuracy benchmarking, and multi-tenant filtered search scenarios.

A deep dive into the practical techniques for reducing the memory and compute footprint of high-dimensional vectors in production search systems.

MongoDB Atlas Vector Search and Dedicated Search Nodes reach GA, providing dedicated infrastructure to isolate search workloads from database operations for improved performance and scalability.