About Me
I'm an undergraduate computer science student at the University of Massachusetts Amherst graduating in May 2026. I'm advised by Prof. Shlomo Zilberstein at the Resource-Bounded Reasoning lab, where I currently research multi-agent Vision-Language-Action coordination. I'm also a part-time SPAR Research Fellow, evaluating LLM alignment targets under Andy Liu's mentorship. I've worked at Nasdaq for 3 terms, building and deploying several multi-agent LLM systems. I was also a Project Lead at BUILD UMass, where I led the development of an LLM system for a non-profit, New England First Amendment Coalition. My personal projects span from theoretical explorations in developmental interpretability to full-stack LLM applications.
I enjoy building projects that allow us to better understand models or that allow models to better understand the world. I'm particularly interested in AI alignment, interpretability, and physical intelligence (VLAs, world models, video-action models, etc...).
Education
Research
- Evaluating and improving LLM alignment targets
- Training Vision-Language-Action models and designing coordination frameworks for multi-agent systems
- Developing MARL and data generation environments in Isaac Lab for large-scale HPC parallel policy training
- Trained, evaluated, and fine-tuned neural networks for animal pose-estimation and behavior classification
Experience
- Developed a fully deployed LangGraph pipeline with safety guardrails, serving over 5,000 Nasdaq-listed companies
- Implemented web, SQL, Elasticsearch, and deep agents on high-stakes sustainability data from 11,000 companies
- Built a multi-agent system from concept to deployment for streamlining the product development cycle
- Engineered LLM orchestrator agents for multi-step decision-making, saving ~37,500 hours ($2M) annually
- Automated document generation and processing using LLMs, Python, LangChain, AWS, and React
- Reduced processing time 10x, contributing to an estimated $15.4M in annual savings
Leadership
- Led a team of 13 SDEs and PMs on an initiative with New England First Amendment Coalition (NEFAC)
- Developed a multi-agent production-ready RAG system using LangChain, FastAPI, AWS, React, and Python
Featured Projects
Saruman
Python
An AI red teaming tool that simulates attacks to benchmark LLM security. Features a variety of attacker personas, configurable defenses, real-time conversation streaming, and experiment analytics. Built with FastAPI + React + LiteLLM.
SLT Experiments
Jupyter Notebook
Singular Learning Theory research experiments in the context of curriculum learning, finetuning, and direct training in modular arithmetic tasks using PyTorch and devinterp.
Franka Table
Python
A simulation environment of 4 Franka Panda arms on a table for multi-agent RL training using MuJoCo and Gymnasium.
Spot Multi-Agent
Python
An Isaac Sim environment for the Spot robot with arm, designed to train multiple VLAs.
TinyLLMs
Python, JAX
Small JAX experiments on a character-level language model, comparing constant, linear, MLP, and two-layer baselines.
Precision Leukemia Treatment
Python
Developed 8 ML models on clinical data of 4,600+ patients to predict transplant outcomes with feature importance analysis.
EliteCode
TypeScript, Swift
Co-founded an educational micro-learning interview-prep mobile platform with serverless AWS backend. Left to pursue AI research.