Themis AI

Cambridge, Massachusetts, USA · 11-50 employees · themisai.io

Themis AI helps organizations build AI systems that make reliable decisions. We combine proprietary decision technology with deep technical expertise to help models and agents perform better, act reliably, and improve as conditions change. Our decision-layer platform connects model-level risk signals with policies, evidence, and operating context to determine what should happen next: proceed, verify, gather more evidence, use another model or tool, or bring in an expert. This helps teams automate routine cases and focus review where it can change the outcome. Underpinning this work is Capsa, our technology for adding uncertainty-estimation capabilities to AI models. Capsa modifies model architectures and derives risk signals from internal model behavior to help identify unfamiliar inputs and unreliable outputs. These signals support both immediate decisions and targeted model improvement, including choosing which examples warrant expert labeling. We work across engineering and operational workflows: from deciding when another simulation or measurement is worthwhile to checking extracted information, reconciling records, and assessing proposed agent actions. Founded at MIT in 2021 by Prof. Daniela Rus, Dr. Alexander Amini, and Elaheh Ahmadi, Themis AI is an MIT Computer Science and Artificial Intelligence Laboratory spinoff built on more than a decade of foundational research. If errors are expensive or expert review is limiting your AI workflow, we’d welcome a conversation.