Vishnu Vardhan

At Mercor, I evaluate AI-generated code changes against live application state using RESTful APIs and MySQL-stored snapshots. I independently verify task completion and identify false success claims, incomplete responses, and other model failure modes.

I also create and refine evaluation scenarios, prompts, and test data, tracking accuracy, relevance, completeness, consistency, and instruction-following to support reliable AI product assessment.

At Regions Bank, I engineered Java Spring Boot microservices for account servicing and case update APIs. API and integration testing reduced UI integration defects by 24%, while GitHub Actions pipelines with SonarQube scans and JUnit tests helped raise backend test coverage to 82%.

In my projects, I developed Python FastAPI tools integrated with OpenAI API and LangChain for pull-request reviews, and built a prototype that analyzed site photos for field inspections. I also implemented state-based verification and responsible AI controls, including action allowlisting, risk classification, and audit-trail capture.