Models that decide
We’re building DecGen (decision generation) models: models that decide instead of generate. We’re starting with customer support agents.
Starting with customer support
Support agents make decisions all day: which order, which policy, what to tell the customer. DecGen models are built for that work.
Faster replies
Customers get answers sooner.
Lower cost
Less compute on every conversation.
More reliable
Fewer mistakes your team has to clean up.
How we work with you
- 01
A 15-minute call
Where your support agent is slow, costly or unreliable.
- 02
You share examples
Tool schemas and sample conversations. NDA if you want.
- 03
Fit to your workflow
A DecGen model built around your tools and policies.
Built by
-
Shrey Vishen (LinkedIn, opens in a new tab)
Inference research at Berkeley’s Sky Lab, running a 122B model on a 36 GB Mac. Before that, ML at Emergent (YC S24).
-
Jason He (LinkedIn, opens in a new tab)
Built AI at two YC S25 startups, Frizzle and Relling. Inference research at Berkeley’s Sky Lab.
Models should decide, not write, their way through the world
We think every model will work this way. Customer support is where we start.