Envariant

Interpretability and reasoning infra for foundation models.

Y CombinatorWinter 2026ActiveB2B

What they do

Problem

Foundation model builders lack clear visibility and control over model reasoning and behaviors, making it hard to analyze, steer, and ensure safe, reliable outputs.

Solution

An AI interpretability and reasoning SDK that provides infrastructure to analyze, steer, and control foundation model behaviors.

Who it's for

Foundation model builders and AI developers at companies building and deploying large models.

Founders

Varun Agarwal

Founder

Working on interpretability infrastructure for foundation models! My background is in AI and bioengineering research at places like Stanford, MIT, Inceptive, and NASA.