CV

Applied Scientist II Intern (L5)

Amazon

Optimized large language model (LLM) performance for real-time applications in New York City, NY.

• Delivered scalable knowledge distillation approach for internal LLM application, improving inference latency by 4.5x while maintaining accuracy
• Collaborated with cross-functional teams to deploy distilled models into production at scale
• Developed internal artifacts for vLLM deployments, Supervised Fine-Tuning (SFT), and Reinforcement Learning with Human Feedback (RLHF)

Research Assistant

University of New Hampshire

Researching robust offline and online decision making under uncertainty. Developing algorithms for complex and involved domains.

• Published in the first ever RL Conference 2024.

Bachelors

University of New Hampshire

Computer Science

• Reinforcement Learning
• Machine Learning
• Deep Learning

Graduated Summa Cum Laude

University of New Hampshire

Awarded to students who are in the top of their graduating class academically.

Deans Scholarship

University of New Hampshire

Scholarship awarded to students with exceptional academic performance, scholarly excellence, and leadership qualities.

Phillips Family Engineering Scholarship

University of New Hampshire

Awarded to Engineering students who maintain a 3.5 GPA or higher in the first 2 years of study.

Computing Monetary Risk Measures in Linear Time

Developed a linear-time algorithm for computing monetary risk measures, which are widely used in reinforcement learning and finance, optimizing state-of-the-art methods and achieving a log-factor speedup in computation time.

Risk-Averse Total-Reward Reinforcement Learning

NeurIPS

Developed risk-sensitive reinforcement learning algorithms that optimize for risk-adjusted future rewards rather than expected reward, giving practical algorithms for optimizing the entropic utility of the total reward, a common risk-sensitive objective in domains such as finance, health, and autonomous driving.

LLM Thematic Analysis for Agriculture Research

Led a collaboration with the UNH Agriculture Dept. to develop a thematic analysis pipeline for categorizing agriculture research laboratories across the USA, implementing a human-in-the-loop LLM system built with LangChain to improve accuracy and reduce workload.

Deep Reinforcement Learning based Optimization of an Island Energy-Water Microgrid System

Resources, Conservation and Recycling

Implemented a Deep RL method to optimize the operation of an island energy-water microgrid system. The results show that the proposed method outperforms the traditional multi-object optimization approach as well as improves the sustainability of the Island infastructure compared to human occupants.

ROIL: Robust Offline Imitation Learning

RLJ (Reinforcement Learning Journal)

Developed novel IRL method to solve for a policy which is maximially robust to adversarial perturbations in the reward function and demonstrations. We show that this approach completely avoids covariate shift, which harshly affects the performance of existing methods.

PPO: Proximal Policy Optimization

Open source project on a flexible implementation of PPO. Implemented in Python. Using Pytorch Lightning.

• Deep RL
• RL
• Python

Bandit Algorithms

Open source project on bandit algorithms. Implemented in Python. Following the popular book by Tor Lattimore and Csaba Szepesvari.

• Bandits
• RL
• Python