Patrick Piper

I like to make things just work.

I'm a senior software engineer at NVIDIA in New York, where I build distributed ML infrastructure for the large multi-modal foundation models behind autonomous vehicles. My work has touched the whole model lifecycle: the curation and enrichment pipelines that feed training, the training and evaluation frameworks that ML engineers build on, and large-scale GPU inference for finding corner cases.

My resume on this site may not be up to date.

From my teams

Cool stuff my work supports. Credit belongs to the named authors, all brilliant engineers, many of whom I get to work with every day.

NVIDIA · Alpamayo ML Infrastructure

• Alpamayo 1: Bridging Reasoning and Action Prediction for Generalizable Autonomous Driving in the Long Tail — Pavone et al., 2025. I grew a legacy dataset pipeline into the daily release system that ships training data for Alpamayo-R1 and other foundation models.
• Scaling PyTorch Dataloaders 10x: A Case Study — Stenbit, Doblar, and Cheng, PyTorch Conference North America 2026. My release pipeline produces the clip data this dataloader reads. I worked with the authors on their feedback to support the per-clip packed format the talk describes.

Outside of work

• I've been playing a lot of Magic: The Gathering lately. Here are some of my unmanicured Commander decks.
• I used to raid with a progression guild in World of Warcraft Classic across two expansions. Our Icecrown Citadel speed run ranked in the top 125 worldwide, and we finished every Cataclysm raid in the top 75 in North America.