1 min read
Add as a preferred source on Google

Sqwish Raises $3 Million to Build Its Tunable Decision Model

Customers can use production data to adjust kettei fat’s judgment. CEO Ushnish Sengupta says the tuning can continue as their tasks evolve.

A brass selector rests among three matte ivory tiles. / TokenPost.ai
A brass selector rests among three matte ivory tiles. / TokenPost.ai

Sqwish, a U.K.-based AI startup, raised $3 million in a pre-seed round announced Tuesday as it develops a decision model customers can tune with their own production data.

Its kettei fat model selects from preset options and records its confidence without generating output tokens. The format is designed for production use cases where a quick selection can matter more than the reasoning offered by a slower, more expensive frontier model.

Sqwish’s built-in post-training tools let customers use production data to adjust the model’s judgment. CEO Ushnish Sengupta said customers can keep adding data so “their models can also keep getting better at the tasks that they care about.”

Sengupta described benchmark scores as “an invitation to try the model out,” with ongoing tuning as the reason customers may stay. Developers can compare models in their own settings by swapping one for another and measuring performance on a specific task.

Sengupta and co-founder Federica Freddi met while graduate students at Cambridge. Sqwish had been developing small language models before shifting its architecture toward decision-model use.

John Kim

John Kim reports on the digital-asset business for TokenPost. Send corrections or tips to info@tokenpost.com.

Loading…