MongoDB Previews Atlas Vector Search, Releases Relational Migrator

MongoDB is looking to make it easier for developers to build applications with generative AI with the introduction of MongoDB Atlas Vector Search. Vector Search, which enables semantic search, is among several new capabilities coming to the MongoDB Atlas developer platform revealed during last week’s MongoDB.local NYC conference in New York.

Also coming to MongoDB Atlas is real-time stream processing for responsive and event-driven applications. In addition, the company announced the general availability of the MongoDB Relational Migrator tool it previewed last year, which migrates data in Oracle, Microsoft SQL Server, MySQL and Postgre SQL databases to any MongoDB deployment including Atlas.

MongoDB Atlas is the company’s multicloud deployment platform for developers consisting of cloud database and database services. The Atlas platform is built on the document model of the MongoDB cloud database that maps data to objects and code.

Real-Time Stream Processing

MongoDB Atlas Stream Processing, available in private preview, promises to accelerate real-time data streaming in the MongoDB database.

“Processing streaming data is hard. It’s complex, is challenging, and it becomes even harder using rigid and inflexible schemas, such as relational platforms,” MongoDB CEO Dev Ittycheria (pictured above) said during the event’s keynote session.

With MongoDB Atlas Stream Processing, “using MongoDB is going to be easy to analyze streaming data and leverage in your applications,” Ittycheria added. “And with MongoDB you can truly unify data in motion with data at rest, so this will transform the way developers build event-driven applications.”

Embedding Meaning Into Atlas Vector Search

The new MongoDB Atlas Vector Search feature, also introduced by Ittycheria, is now available in public preview.

“We will be embedding vector search into our core platform,” he said. “With the addition of vector search, there’s only one platform that unifies source data, metadata, search indexes and vectors in a unified and elegant way.”

Sahir Azam, MongoDB’s chief product officer, emphasized that vectors are a core semantic search component.

MongoDB’s Sahir Azam walks the audience through Atlas Vector Search at the company’s event in New York City.

“Vectors are a powerful tool for working with data because they can represent real-world entities,” Azam said. “A song, an image, a video, a poem can all be vectorized as points in an ‘n-dimensional space. The various dimensions of a vector describe their characteristics, their meaning, so you can relate things that are similar. And, of course, you want to be able to query that in unique ways in a very elegant fashion. It’s a foundational element that can be applied to a wide variety of use cases.”

MongoDB has worked with several dozen design partners on these capabilities, Azam noted. Among them is identity management provider Okta, which is working with MongoDB to enable automatic authentication of database credentials. Another design partner is a startup building a call transcript solution that can identify customer sentiment.

Andrew Davidson, MongoDB senior VP of products, told Channel Futures that developers now can encode meaning into a vector.

“With Atlas Vector Search, we can find similar meaning from other data that we’ve encoded and vectors in the database,” Davidson said.

Enabling Semantic Search

Davidson explained that vector search could enable various use cases typically described as semantic search, such as looking at multiple call transcripts on a single topic or images of people who look similar.

MongoDB’s Andrew Davidson

“All of this has gone into overdrive over the last six months with the rise of large language models and generative AI,” Davidson said. “If you look at the common workflows for building those types of applications, they involve using a vector search database to essentially take the input from the person who’s trying to ask for help from, say, a chatbot. You then send what the person is asking for into a vector search to find similar relevant contexts from your own unique business domain or knowledge base. And then you can prompt it back through the LLM and return the result in a cogent answer to the user.”

According to Davidson, machines must evolve access patterns to let developers build applications that change more rapidly than ever and that can perform at scale.

“We’re just at the beginning right now,” he said.

Currently, human-to-LLM style interfaces are slow because they are single-layer, but Davidson said that more sophisticated multilayered machine-to-machine type applications will appear over time.

“All of this will lead to performance and scale demands like we’ve never seen before,” he said. “And there’s going to be interest in being able to take advantage of the best AI building blocks from each of the hyperscalers.”

Relational Migrator

MongoDB’s Jeff Sposetti

MongoDB introduced its Relational Migrator a year ago at its New York City event, then known as MongoDB World. The tool, available now as a free download, aims to analyze relational schema and suggest how to map it to MongoDB schema. MongoDB designed its user interface to consolidate large volumes of tables into fewer collections with embedded documents.

“Developers can point to their relational database, look at the schema they have, and show a default document schema,” said Jeff Sposetti, MongoDB VP of product management. “It then makes some recommendations on how to migrate the data and to do the transformations.”