RelationalAI and Snowflake join forces to revolutionize enterprise AI decision-making

RelationalAI, a Berkeley, California-based artificial intelligence (AI) startup, announced today the release of a product it’s calling an AI “coprocessor” built for Snowflake, the popular cloud data warehouse provider. The coprocessor integrates relational knowledge graphs and composite AI capabilities into Snowflake’s data management platform. The startup announced its preview availability at Snowflake Summit 2023, an annual user conference.

The new offering underscores Snowflake’s push to become an end-to-end platform for enterprise AI and RelationalAI’s vision for an integrated approach to building intelligent applications. “We’re bringing the support for those workloads inside Snowflake,” RelationalAI CEO Molham Aref said in an interview with VentureBeat. “In the same way a knowledge graph makes it easier for a human to know what’s going on in the data, it makes it easier for a language model.”

Aref explained how RelationalAI integrates with data clouds and language models, and how it enables customers to build knowledge graphs and semantic layers on top of their data.

The coprocessor allows Snowflake customers to run knowledge graphs, prescriptive analytics and rules engines within Snowflake. This eliminates the need to move data out of Snowflake into separate systems for those capabilities. Customers can now build fraud detection, supply chain optimization and other AI-driven applications entirely within Snowflake.

Empowering enterprises with better data

RelationalAI’s AI coprocessor can run securely in the Data Cloud with Snowpark Container Services, a new feature that Snowflake announced at this week’s summit. Snowpark Container Services allows customers to run third-party software and applications within their Snowflake account, enhancing the value of their data without compromising its security.

RelationalAI has demonstrated impressive early adoption across industries including financial services, retail and telecommunications. Several notable organizations are using RelationalAI for business-critical workloads in production today.

“The amazing thing about language models is, you can ask them general questions, and often they can just answer from their internal references,” Aref told VentureBeat. “Sometimes you might ask questions like, ‘How much money did this telco lose due to fraud last year?’ A language model has never seen [the company’s] cost data or financials. So it can’t answer that question. But if you can point it to where [the company’s] data lives, and you ask it, and it can translate from that question to SQL queries, it will be able to give you the answer to that question.”

“So how do you get language models to talk to databases?,” he asked. “Well, one way to do it is to get them to talk directly to databases, which is fine. It works some of the time. But if you have 180 million columns worth of information, that’s more likely to confuse the language model. So what a knowledge graph lets you do is actually build a semantic layer on top of all these data assets. The knowledge graph makes it easier for a human to know what’s going on in the data. It makes it easier for a language model because the language model is trained on text that humans wrote and sort of understands the world in the same way that we understand it using the same terminology.”

The future of data clouds and relational knowledge graphs

Aref also shared his vision for the future of computing with the combination of language models, data clouds and relational knowledge graphs.

“I really think those are the three legs of the stool — they’re going to be at the core of every platform for building decision intelligence in the enterprise,” he said. “Knowledge graphs are central to making it all work because they provide a simplifying abstraction that makes it possible for things to talk to each other. So it’s a very important kind of connection point between language models and humans and databases. So it gives us a common language to talk to each other with.”

RelationalAI is one of the few startups that are tackling the challenge of building intelligent applications with composite AI workloads. The company was founded in 2017 by Aref, who has a background in AI, databases and enterprise software. The company has raised $122 million in funding from investors such as Addition, Madrona Venture Group, Menlo Ventures, Tiger Global and former Snowflake CEO Bob Muglia.

Also a board member at RelationalAI, Muglia praised the company’s technology and vision in a press release.

“The emergence of language models has completely changed the computing landscape,” Muglia said. “As transformative as language models are, their effectiveness can be further amplified when combined with cloud platforms and relational knowledge graphs. I believe this combination will define the future of computing, unlocking powerful capabilities and giving organizations new superpowers.”