1 articles with this tag
Envariant (YC W2026) is building an interpretability SDK that operates inside foundation models, at the level of activations and latent representations, to detect, trace, and steer model behaviors before they become visible failures. It's what the AI observability stack has been missing.