CV

Advisor: Sanmay Das

• Built and validated models of real-world human decision-making using large, noisy administrative datasets; emphasized replicable pipelines, clear error analysis, and decision-relevant metrics.
• Developed interpretable ML baselines (short decision trees) to approximate “default” judgment policies and quantified systematic deviations; used this framework to study consistency and outcome correlates of overrides.
• Designed evaluation protocols for AI judgment systems (including repeated-run robustness checks, aggregation from pairwise preferences to global rankings, and structured failure-mode analysis).
• Implemented simulation and sequential decision-making methods (bandits / learning dynamics) relevant to adaptive sampling strategies and resource allocation under uncertainty.
• Produced well-documented Python tooling for data preparation, feature generation, model training, and reporting to support iterative experimentation with strict temporal separation.

• Supported instruction for graduate ML (office hours, grading, exam logistics); strengthened communication of technical ideas and careful evaluation practice.

Developed a dynamic model of costly, limited human intervention; characterized threshold-style policies and how distributional properties shape when to “spend” discretion under capacity constraints.

Designed and evaluated AI judgment pipelines with robustness checks (run-to-run consistency), alignment tests against established metrics, and clear communication of failure modes for operational settings.

Built short, interpretable models of “default” human decision heuristics and quantified systematic deviations; analyzed consistency and when deviations are associated with improved outcomes—transferable for validating human scoring behavior.

CGPA: 4.0/4.0

CGPA: 3.96/4.0

CGPA: 3.82/4.0