I am a cofounder of OMGrab, where we build smartglasses and data infrastructure to collect massive scale robot training data. Our work focuses on closing the training data gap between robotics and language by making it easier to collect human POV demonstrations. We make smartglasses containing robotics grade cameras, and build data infrastructure for large model training.
I spent several years working on real-time control tasks at Agentic (now part of Epic Games), where we trained small machine learning models to play video games on-device in real time. While at Agentic I also co-authored Enact, a Python library for tracking the output of generative software and agents.
Before joining Agentic I did machine learning research at the Institute for Adaptive and Neural Computation at the University of Edinburgh advised by Sohan Seth. There I published two papers as first author (see below) applying deep learning to population mapping for UNICEF.
Publications
Census-independent population estimation using representation learning
Scientific Reports, Volume 12, Issue 1 (2022)
Towards Sustainable Census Independent Population Estimation in Mozambique
AI for Public Health Workshop @ ICLR'21 (2021)