Meta:

Research Intern, Machine Learning (PhD)  May 2026 - Aug 2026

• Defined the research approach for an open-ended pilot project to select truthful supplemental campaign-performance signals for third-party AI advisers allocating advertiser budgets across publishers.
• Built a market simulator for counterfactual scenarios using literature-based transition equations with parameters calibrated to campaign data; compared simulated and observed trajectory distributions using KS and moment tests. Modeled unknown competitor performance using scenarios of Meta's performance relative to other publishers.
• Built an agentic system that uses multi-round adversarial debate among LLM agents role-playing domain experts to design interpretable decision graphs; trained graph parameters with LLM-guided updates and policy-gradient RL for deterministic, low-latency serving.
• Reduced advertiser budget cuts or increased allocations to Meta in 20% of evaluated simulation scenarios compared with a no-steering baseline.
• Accelerated simulation by ~6x and cut model-serving costs to ~1/20 of baseline through parallelization and caching; the team adopted the simulator and steering-policy outputs for production integration.

Akamai Technologies:

Research Intern, Mapping  Jun 2025 - Aug 2025

• Developed a stochastic CDN traffic simulator to assess machine utilization and tail latency under candidate DNS load-balancing policies, modeling connection arrivals, request timing, and joint distributions of request counts, workloads, and connection lifetimes.
• Characterized five traffic categories from sampled production logs using connection-level features and k-means clustering; analyzed how traffic patterns and extreme workloads affect simulated machine utilization.
• Found agreement between simulated and observed mean machine utilization; evaluated existing load-balancing policies to establish baselines for future policy comparisons.

Aetna at CVS Health:

Data Scientist  Jun 2021 - Jul 2022

• Built a distributed Monte Carlo simulation pipeline to estimate member disenrollment, completing 100K+ simulations on datasets with millions of rows within one hour through parallelization and vectorization.
• Developed optimization and machine-learning methods for experimental design of campaigns, risk-cohort identification, and churn analysis using quadratic programming, Dask, Hive, and supervised learning.
• Built automated data, modeling, and reporting pipelines and developed NLP analyses of sales-call transcripts.

Institute for Health Metrics and Evaluation:

Research Assistant  Feb 2019 - Sep 2019

• Developed statistical and simulation methods, including particle MCMC, for modeling travel flows and improved data-processing and storage pipelines for repeated simulation studies.