Samuel Kahn

CTO & Head of AI, Tickr Senior AI Advisor, ZG Subpoena Solutions Research Fellow, UC Santa Cruz

I lead AI strategy, build AI systems, and grow AI teams. My first AI job, in 2013, was training recurrent networks to forecast microgrid demand at NASA. Today I'm CTO and Head of AI at Tickr, where I lead the team building RiskWise, an agentic legal risk-intelligence platform I’ve led from R&D, to pilot, to a product licensed by major insurers and Big Four firms.

I'm also a Research Fellow at UC Santa Cruz, where I work with scientists to apply generative models to hard scientific problems. One is NEO, a conditional generative model for telescope super-resolution, featured by NVIDIA. Another is a deep-learning model to test scientific theory for NASA's Libera mission.

And as a senior AI advisor to ZG Subpoena Solutions, I build the agentic systems and evaluation harnesses behind automated subpoena compliance and law-enforcement response.

I am also the youngest open sailing world champion ever, with two additional world championship silvers and one bronze, a US National Team member and a Transpacific Yacht Race winner.

I photograph the night sky, which has won some awards along the way. Skiing and climbing are fun too, and most of the rest of my time goes to family.

What I do

Leadership. Build and lead AI organizations, owning strategy, product, and the client, investor, and partner relationships. Grew Tickr's AI and AI-engineering function from founding scientist to fifteen, and serve as RiskWise's technical face with Fortune 10 to Fortune 500 buyers.

Agents and post-training. Post-train open-weight LLMs, build agentic systems, and keep them fast through evaluation and inference optimization. A 4B model I post-trained cuts errors roughly 50% versus GPT-5.2 at nearly 20× lower cost, with no p95 latency penalty.

Production AI systems. Design and scale pipelines that run multi-stage enrichment and LLM inference over large data volumes on distributed infrastructure. Currently moving critical inference off hosted frontier APIs and onto fine-tuned open-weight models on our own GPUs.

Risk intelligence. Build agentic systems that surface latent and emerging risk from messy public and licensed data. A Thompson-sampling bandit over research agents, rewarded in code rather than by a judge model, feeds a model that predicts multidistrict litigation at 0.90 held-out AUC.

Legal AI. Lead legal-AI strategy for three legal-tech startups, two of them acquired, and stay hands-on with the modeling. Raised a legal-process classifier from 78.0% to 97.6% with DSPy, MIPROv2, and Bayesian optimization.

AI for science. Use deep learning and agents for scientific measurement and discovery. First author on NEO (featured by NVIDIA), co-author on a Nature Portfolio paper, and built an agentic discovery system that ran 49 analyses on 25 years of satellite data with an adversarial gate on every finding.