MongoDB recently announced the acquisition of Voyage AI, a startup specializing in generating embedded AI models. The financial details of the transaction have not been disclosed. Prior to this acquisition, Voyage AI had raised $28 million from investors such as Snowflake and Databricks.
MongoDB is a publicly traded company on the Nasdaq stock exchange, offering a widely popular document database also named MongoDB. Unlike traditional relational databases that require all records to follow the same format, MongoDB's document model offers more flexibility, simplifying application development processes.
MongoDB provides various versions of its database, including MongoDB Atlas, a managed cloud version. This version eliminates the need for customers to manage the infrastructure required to run the database and streamlines maintenance tasks, particularly patch updates and query optimizations to reduce hardware usage.
MongoDB Atlas supports numerous use cases, including powering AI applications. The acquisition of Voyage AI aims to enhance these capabilities.
When AI models process business documents, they do not handle raw formats directly but convert them into embedded structures first. These embedded structures are mathematical representations that store key information about the file and describe its relationship with other records in the database. Embedded structures are typically generated by AI models.
Voyage AI has developed several AI models for generating embeddings, including the flagship algorithm voyage-3-large launched last month. In terms of embedding quality, this algorithm claims to outperform competitors from OpenAI and Cohere Inc. by 9.7% and 20.7%, respectively.
The acquired company also offers specialized models for various verticals, optimized to generate embeddings from code files, legal documents, financial data, and more.
Voyage AI also provides re-ranker models that can reorder search results to display the most relevant items. When AI applications retrieve a series of data points to answer prompts, re-rankers are used to identify the most pertinent data points.
MongoDB plans to integrate Voyage AI's models into MongoDB Atlas later this year. Once integrated, developers will be able to quickly convert records into embeddings and store them within the database. AI applications built on Atlas will leverage Voyage AI's re-rankers to find the most relevant data for a given task.
MongoDB believes this integration will simplify the application development workflow. Typically, software teams keep standard data and embeddings for AI workloads in two separate databases. Voyage AI's technology will make it more practical to store both types of data in Atlas, thereby simplifying coding workflows.
Following the integration, MongoDB also intends to enhance the functionality of Voyage AI's models in Atlas. The company plans to support multi-modal data like images and videos and introduce more specific features for industries such as law and finance.
