Through contract work for Turing, Pareto, and ValsAI, I evaluate and benchmark LLMs and AI coding agents. I’ve engineered Harbor pipelines for documented CVE tasks, checking that vulnerable code is reachable and that issues reproduce consistently.
For Taiga, I authored prompts with reference solutions and designed rubrics to compare models against the same criteria. On Terminal-Bench, I containerized tasks, ran baseline agents, and validated candidate patches against test cases.
I also conducted Turing RLHF feedback loops and adversarial safety testing, documenting model failure modes for review. I debugged inconsistent test cases to distinguish model failures from problems in the tasks or environments.
Previously, I engineered and maintained Ruby on Rails services at Machinetools, including OpenAI-powered theme generation and GraphQL data synchronization. At Arbisoft, I led Python/Django development and built AWS data pipelines; earlier at Devsinc, I delivered full-stack features for client projects.
