Anish Agarwal

Anish Agarwal is a cofounder of Traversal, where his team develops AI site reliability engineer agents to help enterprises diagnose and mitigate production incidents. His research spans causal machine learning, reinforcement learning, and high-dimensional statistics. Columbia’s November 2025 entrepreneurship event biography identified him as Traversal’s CEO and cofounder.[3][5][7]

Agarwal cofounded Traversal with Raaz Dwivedi, Raj Agrawal, and Ahmed Lone. He and his team announced Traversal AI on the Reliable Agents track at the AI Engineer World’s Fair, according to Latent Space’s June 2025 recap. Traversal describes his move into company-building as an opportunity to combine research with commercialization after achieving his goal of becoming a professor.[3][1]

In the description for his Columbia presentation scheduled for November 18, 2025, Agarwal framed incident troubleshooting as a bottleneck to end-to-end automation of the software development lifecycle. He described automated incident response as joining agentic system design, large language model evaluation, causal inference on unstructured data, and time-series modeling of high-dimensional telemetry. The task, in his formulation, is to enable AI systems to “form hypotheses, reason under uncertainty, and propose actionable remediations” despite incomplete information, noisy signals, and strict reliability constraints.[7]

Before Traversal, Agarwal’s research addressed how to use observational data to estimate outcomes under alternative interventions. That earlier work provides context for his expertise in causal inference and data-driven decision-making.[6][5]

Agarwal coauthored Synthetic Interventions with Devavrat Shah and Dennis Shen, first submitted to arXiv on June 13, 2020. The August 2024 version extends synthetic controls—a method used for policy evaluation with data tracking units over time—to multiple treatments. Its low-rank tensor factor model adds latent treatment factors to the unit and time factors used in the conventional matrix model, supporting a generalization of synthetic-control estimators.[6]

Under that model, the authors prove consistency for one instantiation of their approach and give conditions for asymptotic normality. They also study prediction performance in a simulation and revisit the canonical synthetic-control case study of anti-tobacco legislation, exploring related questions beyond the original analysis.[6]

Agarwal earned undergraduate and master’s degrees at Caltech under the supervision of Mani Chandy and Adam Wierman, then worked as a management consultant at Boston Consulting Group before pursuing his doctorate. At MIT EECS, his doctoral advisers were Alberto Abadie, Munther Dahleh, and Devavrat Shah. His personal biography describes a subsequent postdoctoral fellowship at UC Berkeley’s Simons Institute followed by a research scientist role at Amazon Core AI. It also records time at Microsoft Research and technical consulting for TauRx Therapeutics and Uber on experiment design and causal inference. Columbia’s faculty directory lists him as an assistant professor of Industrial Engineering and Operations Research.[5][2]