Projects

Advised by Sayash Kapoor at the Science of Agent Evaluation Lab (SAgE). Working on automated agent error detection and optimization via GEPA.

Advised by Prof. Aarti Gupta. Improving DSPy coding agent SWE-Bench-Lite performance with GEPA. Achieved 44% Pass@1 success rate for Qwen 3.0 Coder 30B A3B.

Explored data-efficient fine-tuning of LLMs by applying LESS data selection and QLoRA. Fine-tuned Llama-2-7B on single-task math with just 1k prompts.

Built compilers by following the Making Compilers book. Projects include tree parsers, type checking, SSA-based optimizations, and recreating a C-like compiler with a working runtime.

Led development of a production-ready analytics tool for a New Jersey food pantry. Built with Vue, TypeScript, and Tailwind, delivering actionable insights for non-profit operations.