At Visa Inc., I engineered R2D2, a production AI software agent that turns Jira tickets into CI-gated pull requests. It reached 800+ engineers and generated 12K PRs across 450 repositories in six months.
I also architected a hybrid retrieval system across 8.4M lines of Python, improving Recall@10 from 0.46 to 0.89. I built safety and evaluation gates that blocked 150–200 insecure or policy-violating suggestions monthly.
I created Lucin, an open-source static security analyzer for AI agents, which reached 5,000+ installs in its first month. Its analysis found live data-exfiltration paths in 114 popular open-source repositories, and I shipped 30 detectors across eight agent frameworks.
At WorldQuant, I developed predictive alphas through statistical feature engineering, backtesting, and out-of-sample validation; 48 alphas above Sharpe 1.5 were accepted into the production pool. I also extended LPython as a Compiler Developer with the Python Software Foundation, and authored two AI papers under review at AAAI 2027 and ICLR 2027.
