projects
LLM Fractional Quantization-- Python, PyTorch, LLMs, Quantization
Research and implementation of fractional quantization techniques for Large Language Models.
Balatro RL-- Python, Reinforcement Learning, PyTorch
Reinforcement learning agent trained to play the game Balatro.
CPS572 Sp26 Mini Project-- Python, LLM, HuggingFace, Tinker API
Multi-task LoRA fine-tuning of Llama-3.1-8B on IFEval, GSM8K, and HumanEval via the Tinker API. Four-stage training: SFT on task-aligned Tulu-3 subsets, mixed RLVR, MBPP RLVR for code, then AllenAI IFEval-only RLVR.
Codebase Tutorial Generator-- Python
AI-powered tool to automatically generate tutorials and documentation from codebases.
Duke Advisor Agent-- Python, ReAct Framework, Tool-Use, Hierarchical Agents, State Management
Production-ready academic advising AI built with a ReAct loop to solve complex multi-step student queries. Features an adaptive planning system, intelligent tool orchestration, and a dual-LLM architecture with persistent context accumulation.
Outpilot.ai-- Golang, Next.js, TypeScript, N8n, Redis, PostgreSQL, AWS
Full-stack SaaS platform for automated lead generation with webhook-driven workflows, scalable job orchestration system, and multi-tier subscription management using DodoPayments.
FuturePreneurs-- Next.js 13, Phaser.js, Express.js, MongoDB, AWS
Platform for E-Cell VIT's flagship event hosting 2200+ participants with 2D real-time RPG game, dynamic quiz system, and automated team management.