Hi, it’s Fortune’s tech fellow Andrea Guzman filling in for David.
Last month, Google challenger and search engine Neeva said it was ending its consumer business after struggling to attract new users.
Founders Sridhar Ramaswamy and Vivek Raghunathan had set out to reinvent the search experience with a new generative A.I. engine, but then it became clear that Microsoft planned to spend billions doing just that with Bing.
“Startups don't really do that well in a world in which there are people that are just outspending it by many orders of magnitude,” Ramaswamy said on a call with Fortune. “That's when Vivek and I decided that we would be better off applying our skills in infrastructure, in search, in A.I. and language models to more of an enterprise setting.”
Now, after being acquired by cloud giant Snowflake, Ramaswamy has plans to build a product that leverages generative A.I. to build SQL queries. Neeva showed demos of it at Snowflake’s summit this week. Ramaswamy said he hopes to launch it in a private preview for customers in a few months.
But even in the enterprise space, Neeva is still guaranteed to come across challengers like Salesforce when aiming to deliver enterprise-ready A.I.
However, Ramaswamy, Google’s former ad boss who has worked in search pretty much his whole life, thinks that Neeva is uniquely positioned. “We are the gold standard in terms of being a safe, secure, trusted, and efficient platform for both data and applications.” He said the team is gathering momentum for what’s called “native applications,” which allow third parties to write applications on top of Snowflake’s, causing Neeva to consider avenues from a first-party perspective and encouraging an ecosystem of developers.
Ramaswamy said that for business use cases, Neeva is aiming to be reliable and predictable. But the tech is early enough that a human will still need to verify the outputs.
Neeva's technique involves typing a question into a chatbot, which will then "retrieve a set of documents or a set of facts that are relevant to the question, supply them to the language model, and then have the model generate answers within the context of the facts that have been provided.”
The purpose of that method, Ramaswamy said, is to make the output more grounded and incorporate real-time data into it.
That emphasis on safety and accuracy has been a long-held value for Ramaswamy. He said growing up, his dad was picky about which newspaper the household bought and avoided those he felt were sensational and that’s driven his values around safety in the industry.
“I'm also old enough to remember the time of cycling to a library to look up an encyclopedia,” Ramaswamy said. “So there's that little kid in me that sort of likes the idea of like, what is trustworthy?”
More news below.
Want to send thoughts or suggestions to Data Sheet? Drop a line here.
Andrea Guzman
