Projects

NVIDIA Collective Communications Library (NCCL)

Optimized Multi-GPU Primitives

• New Features, Communication Kernels, and Protocols (v2.32+)

DSL-Monkeys

Self-Generated In-Context Examples for GPU DSL Kernels

• 2026 ICLR DATA-FM Workshop
• 2026 ICLR TTU Workshop

ParallelKernelBench

A benchmark for evaluating how well AI can write multi-GPU kernels!

• ICML 2026 Deep Learning 4 Code

Links

• Talk
• GitHub
• Blog post
• Hugging Face (problems)
• Paper (alphaXiv)

KernelBench Extended DSL Support

Added a framework to support new DSLs (e.g. ThunderKittens, CuTe, TileLang, etc.) within the KernelBench codebase.

NVSHMEM 3.6.5 Library Development

As part of my 2025 internship at NVIDIA, I added support to Perplexity's MoE Kernels for new nvshmem python bindings. Also added an experimental LTOIR build feature to NVSHMEM.

• pplx-kernels Pull Request
• NVSHMEM 3.6.5 LTOIR Support

Spellr

Arduino-based transcription attachment for pens.

• Pear x OpenAI Hackathon Top 8 Finalist (2024)

PaperCraft

Developed a VR RTS game that lets you direct troops on a live map.

• StanfordVR Hackathon 1st Place (2022)