Sunnyvale-based Brevian wants to make it easy for business users to build custom AI agents. Currently, the company is focused on support teams and security analysts (areas where both use cases and training sets are clearly defined), but plans to expand into other areas over time. The company came out of stealth today and announced a $9 million seed funding round.
Brevian, BREV/ΛN, was founded by Vinay Wagh (CEO) and Ram Swaminathan (CTO). The two came into this startup journey from very different directions. Mr. Wagh previously served as Product Director at Databricks and Head of Product at Bracket Computing. During his time at Databricks, he prepared the company for sale to large enterprises and worked on launching his serverless product, his Databricks SQL Serverless. “It went public in February, and I was watching everything that was going on,” Wagg said. We quickly agreed that we would need all of these missing pieces. There is a lot of hope, but unless someone solves this part… , it doesn't work well. That was the inspiration to go out. Something is happening. And then I met Ram. [Swaminathan] Through a mutual friend. ”
While Wagh has worked a lot on the data side, Swaminathan's background is in the computer science and machine learning side. After working in academia for a while, he spent nine years at Bell Labs and 17 years at HP Labs, working on everything from coding theory to cryptography to early machine learning efforts. He then moved to LinkedIn to lead the AI Trust team, leaving the company in late 2022 to join Wagh in founding Brevian as CTO.
Brevian founders Vinay Wagh (left) and Ram Swaminathan (right).
“I had no product sense at all,” he told me. “You write papers, you write patents, but can you really build a product that impacts a lot of people? And LinkedIn was the perfect target. That's it. It has a huge impact on society. I really enjoy working with product people and engineering and coordinating all of this to create value.”
The team argues that security is at the core of any corporate project around generative AI. For example, AI vendors must ensure that personal information is not compromised. When the company launched, many companies decided to ban his ChatGPT within their organizations because there was no way to guarantee that employees would not enter sensitive data into chat prompts.
In the early days, Brevian was primarily focused on security. “We quickly developed a model that looked like this: [personally identifiable information]” Wag explained. “We realized that this prompt injection attack vector was on the rise, so we built an intent-based system to detect prompt injections. So what we ended up doing at that time was […] I had collected all the misalignment prompts, so I ran the model on all of them and it captured all of them. ”
Image credit: Brebien
Interestingly, Wagg says that even today, companies still have a lot of concerns about security related to large language models, but they don't really know what they're really worried about beyond these leaks. , he pointed out. The team quickly realized that the real challenge was not security issues that were holding back enterprise adoption, but the challenge of building systems that solved real problems.
“Our vision was to enable business users within the enterprise to use AI to simplify their daily tasks. All of this fits into our vision of extending from security to building AI agents. Masu.”
Brevian's seed round was led by Felix's partner Jake Storm (and this is Felix's biggest seed round check to date). He pointed out that a year ago everyone was talking about AI infrastructure, but few were talking about the importance of applications and their security. “2023 is, in a sense, [the year of ]AI infrastructure. 2024 is the year of AI apps. And we felt that AI apps were way ahead,” he told me.
With this funding round, Brevian aims to accelerate product development and expand its team to meet growing customer demand through an early release program.