At DevRev, via MountBlue Technologies, I analyzed AI agent traces to identify failures such as wrong tool choices, incorrect parameters, made-up data, and incomplete retrieval. I also contributed to Enterprise-Bench and created a reusable skill that posts standardized failure summaries, pass@10 accuracy, and verdicts.
I implemented dataset integrity checks and a GitHub Actions workflow for benchmark data changes, which the AI benchmarking team adopted. I also integrated Microsoft Copilot and Salesforce Agentforce into benchmark runs and compared their tool choices and data retrieval.
At Rexera, via MountBlue Technologies, I developed Python Playwright automation for critical web workflows and built a weekly regression suite to validate functionality after deployments. In a personal project, I built an HR policy RAG assistant with hybrid search and evaluated its retrieval and answers using RAGAS.
