Projects

Reimplemented the famous "Show, Attend and Tell" paper in PyTorch and extended it with adaptive attention (λ-tuned sentinel gating) and VQA. Analyzed the tradeoffs of soft attention mechanisms for vision-language tasks and presented poster.

Cornell CS 4782: Deep Learning · team of 3 · May 2026

A compiler that takes the straggler-aware AllReduce schedule from "Efficient AllReduce with Stragglers" and compiles it into executable CUDA/MPI/NCCL code, making the theoretical algorithm runnable on real GPU clusters.

Cornell Data Science · team of 5 · Apr 2026

An interactive demo on Variational Autoencoders, trained using Fashion-MNIST dataset to visualize latent space. Include sections on architecture, latent space demo, and KL divergence explanation, and made in Cornell's CS 4782: Deep Learning course.

Cornell CS 4782: Deep Learning · individual · Apr 2026

A trading backtester built with a Cornell Data Science team sponsored by Millennium's Systematic Data Platform team. Extended with short-selling functionality and realistic trading costs (ADV caps, borrow costs, market impact). Used to implement and backtest a Customer-Supplier Momentum strategy.

Cornell Data Science × Millennium · team of 10 · Mar 2026

A RAG system that indexes Cornell Data Science's GitHub repositories to answer deep technical questions about past projects, with a focus on intelligently chunking and storing context from large codebases.

Cornell Data Science · team of 6 · May 2025

An AI agent system that translates natural language instructions into real-time drone flight commands through a closed-loop planner-control agent architecture.

Emergence AI · with one other intern · Aug 2024

A genetic algorithm that optimizes drone coverage paths under energy constraints, developed as a senior research project at TJHSST.

TJHSST Senior Research Lab · with one lab partner · Mar 2024