At Brio, I design multihop evaluation scenarios for TAU, a benchmark for assessing reasoning and tool use in LLM agents. I also author Structured Prompt Templates that support repeatable task formulation across workflows such as Google Workspace, Jira, and Zendesk.
I apply Structured Prompt Design to create consistent agent instructions, balancing natural-language clarity with API-call precision. I also build and validate gold-standard trajectories for supervised fine-tuning and RLHF-style comparisons.
Previously, as a Software Engineer at Nymble Leap, I developed applications, created data pipelines, and worked on GenAI projects. My work there combined Python development with web development and data science.
At Brio, I also worked as an LLM Training Specialist, training and evaluating models on coding problems based on resolved GitHub issues and merged branches. I’m interested in sports analytics, event-driven architectures, and real-time streaming systems.

