Bellevue-based artificial intelligence startup OpenPipe has raised $6.7 million in seed funding.
With the funding, announced Tuesday, OpenPipe CEO Kyle Corbitt said the company plans to significantly grow its team.
Corbitt co-founded OpenPipe with his brother David Corbitt in 2023. The company fine-tunes large language models, like ChatGPT, to fit the specific needs of its clients. A company trying to find every name mentioned in a certain number of Wikipedia pages, for example, could do so much faster and cheaper through a fine-tuned model, Kyle Corbitt said.
He said the service is live and available to use, but didn't go into specifics about the company's client base. OpenPipe lists Volvo and Rakuten as users and says it has saved customers $7 million this year.
OpenPipe currently has fewer than 10 employees, all of whom are based in the Seattle area, Corbitt said. The company plans to have roughly 20 to 30 employees a year from now.
"I wouldn't want to be larger than that just purely because I think density is pretty important, and it's hard to grow fast," Corbitt said.
Related coverage
OpenPipe has private offices at a coworking space in Factoria. Corbitt said OpenPipe will likely have to expand at some point, though the company doesn't have specific plans right now.
Costanoa Ventures led the seed round. The accelerator Y Combinator participated, along with multiple angel investors. Costanoa has invested in the marketing tech company Amplify.ai and the data company Datalogix.
Kyle Corbitt spent more than four years at Y Combinator as a software engineer and director, according to his LinkedIn page. David Corbitt, meanwhile, brings engineering experience from Palantir and Qualtrics.
Looking ahead, Kyle Corbitt said people are using larger models like ChatGPT right now because the technology is so new, but more fine-tuned models will become popular as the technology evolves.
"We expect we'll be doing more inference than OpenAI. That's our goal, is to be doing the majority of all inference or have models that we've trained doing the majority of inference," Corbitt said.
