I'm a research engineer interested in LLMs, multimodal AI, computer vision, and autonomous systems. I enjoy the research side of ML as much as getting things to work on real hardware.
I wanted to use my Mac mini M4 and NVIDIA laptop together for distributed ML. That turned into TBCCL, an experimental C++ collective communication runtime spanning macOS, Linux, CUDA, and Metal-shared memory, using TCP (including Thunderbolt IP).
Core runtime · PyTorch · vLLM · Exo · Docs
My submission to comma.ai's video compression challenge. It reached #1 on the official leaderboard on October 21, 2026 with a score of 0.134 (lower is better). I worked on rate-aware semantic-token refinement, algebraic renderer folding, and more compact model-state packing.
Code · Submission · Experiment write-up
UniLCD · ECCV 2024
At Boston University's H2X Lab, I worked on UniLCD, using residual reinforcement learning to decide when an autonomous system should compute locally and when it should use the cloud.
Other projects and contributions: Exo · BetterSearch · VidTune · Robotics
Previously a Research Engineer at MBZUAI (Aug 2025 – Aug 2026), working on LLM training, post-training, and distributed inference, from 70B-class fine-tuning to trillion-parameter MoE serving. Before that, I worked on vision-language models and autonomous driving at Boston University's H2X Lab.
Publications: UniLCD · ECCV 2024 · Gait re-identification · ETCCS 2020
I play guitar, enjoy making music, and have spent more time than expected convincing different computers to cooperate. "Hurricane Sandy" was once a stage nickname, thanks partly to some rather spiky hair.
I'm looking for the right research or engineering role. If you're working on something interesting, I'd love to chat.