Souradeep Roy

I've contributed 20+ merged pull requests to Mesa, an agent-based modeling library, improving recurring-event controls, simulation-time safety, caching, and regression coverage.

I built ArgonX, a Bayesian decision engine for A/B experiments that translates posterior inference into ship/no-ship decisions under uncertainty. It combines expected loss, tail risk, ROPE significance, guardrails, and sequential stopping rules, and is published to PyPI with GitHub Actions CI/CD.

In quantitative research, I modeled structural bias in analyst earnings forecasts using hierarchical Bayesian inference and built a leakage-controlled XGBoost pipeline. The model reached 0.536 ROC-AUC and 0.859 PR-AUC using walk-forward validation.

I'm focused on data science, AI/ML, and quantitative research, using Python, probabilistic modeling, machine learning, and robust engineering practices to turn complex data into useful decisions.