Description
MongoDB Vector Search is a powerful feature of MongoDB Atlas that allows users to store and search vectors alongside their operational data. Vectors are mathematical representations of data, expressed as arrays of numbers that capture specific features or properties. In the context of machine learning and artificial intelligence, vector embeddings are used to encode complex data such as text, images, or audio into a format that computers can process. This capability is essential for applications like semantic search, recommendation systems, and generative AI.
One of the key advantages of MongoDB Vector Search is its ability to perform searches based on semantic meaning captured in vectors, as opposed to traditional keyword searches. This allows for more accurate and context-aware search results, enhancing user experience. MongoDB serves as a vector database, enabling users to manage their operational data, metadata, and vector embeddings all in one place, eliminating the need for separate databases and synchronization.
MongoDB Vector Search supports various algorithms for finding similar vectors, including K-Nearest Neighbor (KNN), Approximate Nearest Neighbor (ANN), and Exact Nearest Neighbor (ENN). These algorithms are designed to balance accuracy and performance, making it easier to handle complex and highly dimensional vector use cases. Additionally, the platform supports embeddings from any provider under the 4096-dimension limit, allowing for flexibility in data ingestion and querying.
The integration of native reranking capabilities enhances retrieval accuracy by ensuring that only the most relevant data is delivered to large language models (LLMs), optimizing token expenditure and improving overall performance. Furthermore, MongoDB's hybrid search capability combines lexical and vector search in a single query, providing users with the best of both worlds.
MongoDB Vector Search is suitable for a wide range of applications, including semantic search, retrieval-augmented generation (RAG), and the development of agentic systems. With its robust features and capabilities, MongoDB Vector Search empowers organizations to build intelligent applications that leverage the power of vector data.
MongoDB Vector Search's Core Features
Fully managed, multi-cloud database
Approximate Nearest Neighbor (ANN) search
Exact Nearest Neighbor (ENN) search
Hybrid search capabilities
Native reranking integration
Automated embedding generation
Independent scaling with Search Nodes
Cost efficiency through vector quantization
How to use MongoDB Vector Search?
Create vector embeddings using an open-source model or let MongoDB auto-generate them.
Define a vector search index for the embeddings you want to query.
Use the $vectorSearch stage in the aggregation pipeline for queries.
Combine vector queries with metadata filters for enhanced search results.
MongoDB Vector Search's Use Cases
- Semantic Search
- Retrieval-Augmented Generation
- Agentic Systems
- Recommendation Systems
- Anomaly Detection
