AI startup Braintrust helps developers evaluate and enhance their AI products. The team just raised a $5.1 million seed round from Greylock.

• Braintrust helps software developers test and quickly enhance their enterprise AI applications.
• Founder and CEO Ankur Goyal led the AI team at Figma before starting the company.

Ankur Goyal has been working in AI long before the hype cycle.

The repeat founder started his first AI startup Impira back in 2017, and has worked in machine learning for over a decade, he told Business Insider.

After Impira was acquired by Figma at the beginning of 2023, Goyal joined the company to lead its machine learning platform team. Throughout his time there, though, he kept running into the same problem with generative AI models.

When a large language model is stuck or not operating properly, Goyal explained, it's hard to tell exactly why, and engineers often have to troubleshoot or fine-tune these models by testing lots of inputs to hopefully debug the model.

Instead of going through this tedious process over and over again, Goyal decided to build AI-powered programs that could give software developers a "toolkit" for running evaluations on these models, so they could log, refine, and enhance their AI-enabled products much more quickly.

Since leaving Figma and founding Braintrust this summer, Goyal has raised $5.1 million in seed funding in a round led by Greylock partner Saam Motamedi.

This follows previous pre-seed investments by Elad Gil, Alana Goyal at Basecase, who is Goyal's wife, and several other prominent angel investors including Clem Delangue, CEO of HuggingFace; Greg Brockman of OpenAI; Howie Liu, CEO at Airtable; Jack Altman, CEO of Lattice; and Guillermo Rauch, CEO of Vercel.

This new funding round brings the total capital Braintrust has raised to $8.3 million.

Braintrust helps AI developers by connecting to their cloud workflows and running evaluations on the code they're writing so that it works every time they change the code, Goyal said.

"If we see software engineers integrating Braintrust into their dev loop, they instantly get a sense of what they've improved and what is broken," said Goyal.

While other startups focus on issues of "observability" or detecting errors in their AI workflow, Goyal explained, Braintrust differentiates itself by primarily focusing on "why" something is not working to speed up its repair.

Braintrust also offers tools to help evaluate and clean up company datasets, and test prompts. According to the company website, many of the top AI models are compatible via Braintrust's API, including those developed by OpenAI, Anthropic, Meta, and Mistral.

Many of Braintrust's customers have told them that they built similar internal tools, but that Braintrust's tool was so much more efficient for evaluations that they've stopped maintaining their own and have now fully switched over to Braintrust's, Goyal said.

Some of Braintrust's existing customers include Zapier, Airtable, Coda, and Instacart.