At Coval, I built an AI evaluation platform with Python and TypeScript services and Next.js surfaces. I created golden conversations, calibrated rubrics, and enforced agreement gates that model and prompt releases could not bypass.
I used LangGraph and LangChain to simulate agent scenarios, testing tool use, interruptions, and recovery. I also developed adversarial task suites and converted recurring failures into default graders.
At Constellation Energy, I built retrieval-grounded assistants over tariffs, procedures, and regulatory filings, and created an outage-triage evaluation suite for model selection. I also established model governance and lifecycle documentation in a regulated operator environment.
At Microsoft, I built forecasting and risk models and developed sequential testing and variance-reduction features for the ExP experimentation platform. Earlier at AT&T, I built subscriber and network-performance reporting and automated Python workflows for analyst data.
