July 13, 2024
Digital generated image of abstract AI data chat icons flying over digital surface with codes

When developers have a particular job that AI can solve, it’s not typically as simple as just pointing an LLM at the data. There are other considerations such as cost, speed and accuracy and finding ways to balance all of those has been particularly challenging, especially with so many new models coming online all the time.

That’s where Unify comes in, a British startup from an Imperial College alum, who has come up with a router tool that lets developers enter parameters and find the best LLM for their unique requirements, whatever they may be. On Wednesday, the company announced an $8 million investment.

“The main objective with Unify is figuring out which models from which providers are best for your task using objective benchmarks and dashboards that let you compare them” company founder and CEO Daniel Lenton told TechCrunch.

“The router effectively is kind of a natural extension of this process, particularly as companies start to deploy at scale, and the speed and the cost become more important. So what we’re really trying to do is give people much more control over the quality, cost and speed profile of their LLM applications,” Lenton said.

Unsurprisingly, Unify uses AI to run the core router application. “Our router itself is a learned neural network. So it learns which models are best for doing which tasks,” he said. The company does this by running exhaustive benchmarks on each new model on all these tasks using GPT Pro as a judge. From this, the system learns how good this model was doing certain tasks across their training sets.

“So very quickly any new model provider is supported by the router basically a day or two later,” he said.

Lenton says the router, and how they’ve built a unique model to train it, is in itself a way to defend what his startup is doing from encroachment by the bigger players, that and the fact that they are neutral, and the hyperscalers might not be.

He said typically customers are just experimenting with different models and don’t have a tool to track which one is best.

“There are people that have kind of hair on fire problems that are willing to try an existing solution. So I think that’s how we’ve managed to get our foot in the door,” he said.

While there are competitors out there like Martian Router, OpenRouter and Portkey, Lenton says his company is the only one optimizing jointly for quality, cost and speed.

The company is small right now with seven employees, and he is keeping it intentionally lean while he works on getting a fully monetized product in the market. The plan is to add three additional employees this year.

He reports about 3000 sign ups so far with a few hundred regular users. They expect to make money as they charge companies for building their own custom benchmarks. Each company can get started with the tool with a $50 credit.

The $8 million investment came from a slew of investors including SignalFire, M12, J12, Essence VC, A. Capital. Lunar VC, Y Combinator and a bunch of prominent industry angels.

Source link