AI is transforming industries, but data quality remains a major challenge for truly reliable applications. By acquiring Voyage AI, a specialist in advanced embedding and reranking models, MongoDB enhances its ability to provide precise and optimized information retrieval solutions. This integration aims to limit hallucinations in AI systems and lay the foundation for a new generation of more efficient AI applications.
A strategic advancement for MongoDB and enterprise AI
MongoDB is a document-oriented NoSQL database management system designed to store and manage unstructured and semi-structured data with great flexibility and scalability. Based on a BSON document model (a binary extension of JSON), it allows developers to structure their data without the constraints of traditional relational databases. The company, based in New York, has thousands of customers in over 100 countries and offers a data platform that integrates various services to meet the needs of modern applications.
Thanks to the expertise of its team of researchers from prestigious institutions such as Stanford, MIT, UC Berkeley, and Princeton, the start-up Voyage AI, created in 2023 by Tengyu Ma, has developed advanced solutions to extract meaning from texts and unstructured data, ranging from legal documents to enterprise knowledge bases.
Its integration models improve data accuracy and relevance, thereby reducing hallucination risks. They are used by AI leaders like Anthropic, LangChain, Harvey, and Replit. Its zero-shot models are among the highest rated in the Hugging Face community.
Dev Ittycheria, CEO of MongoDB, explains in a blog post the reasons for acquiring Voyage AI:
"By integrating Voyage AI's extraction capabilities into MongoDB, we help organizations more easily create AI applications with increased accuracy and reliability, without unnecessary complexity."
A three-step integration
Initially, Voyage AI's text, multimodal, and reranking integration models will remain available via their current APIs as well as on AWS Marketplace and Azure Marketplace, ensuring continuity for developers already using these technologies.
In the second phase, MongoDB will gradually integrate Voyage AI into MongoDB Atlas. An automatic integration service for vector search will simplify data management, followed by native reranking to improve result accuracy. In parallel, MongoDB plans to expand its AI capabilities to meet the specific needs of sectors such as finance, law, and code generation.
Finally, the company will strengthen the management of multimodal data (text, images, videos) and introduce instruction models, allowing developers to refine search behavior through simplified prompts.
This three-step approach aims to maximize Voyage AI's capabilities while facilitating the development and optimization of AI applications, thereby reducing technical complexity for businesses.