Enhancing RAG Accuracy: Databricks Ventures Invests in Voyage AI

We consistently hear from our customers that one of the headwinds to transitioning Generative AI applications from pilot to production is the accuracy — or lack thereof — of the results generated by off-the-shelf large language models (LLMs). One of the ways that organizations are addressing this accuracy gap is by building what are known as compound AI systems, which could include a Retrieval Augmented Generation (RAG) architecture. RAG architectures and compound AI systems enhance the quality of responses from off-the-shelf LLMs by incorporating relevant company- or domain-specific data as part of the prompt and response. At Databricks, we view this as a way to shift from LLM general intelligence to what we call data intelligence. Even a small improvement in the quality and efficiency of retrieval can make an outsized impact on the end user experience.

A top-quality embedding model is a cornerstone to an accurate RAG system. And with the explosion of RAG applications developed on Databricks Mosaic AI this year, it is critical that Databricks is able to offer industry-leading embedding and rerank models. That’s why we’re excited to share that Databricks Ventures has invested in the Series A funding round of Voyage AI, a startup co-founded by CEO Tengyu Ma, a pioneer in the field of deep learning.

We’re also announcing a new partnership that brings Voyage rerank and embedding models to Mosaic AI Model Serving for Databricks customers. All of our work with Voyage AI is in service of helping our customers build data intelligence and high-quality production AI systems.

Why Voyage AI

Voyage AI offers some of the world’s best embedding models available today. Their high-quality embedding and rerank models allow companies to enhance RAG search and retrieval accuracy and efficiency, resulting in more accurate RAG solutions and compound AI systems. Their models are also optimized for specific domains and tailored to a company’s data. As a result, Voyage AI’s models are being used by many leading AI companies to deliver best-in-class user experiences.

What’s Next for Voyage AI and Databricks

We will offer Voyage AI’s latest generation of embedding and rerank models natively within the Mosaic AI Model Serving solution. By offering these models natively, we’ll enable our customers to securely and cost-effectively build production-quality RAG applications, and they can do so confidently, knowing that Databricks offers the long-term support and stability that enterprises require.

We expect to announce integration of Voyage AI’s models within Mosaic AI in the coming months. Stay tuned for more information!