About

For over a decade I have taken machine learning projects from research to production in domains as varied as neural interfaces, geophysics, remote sensing, and climate and sustainability.

Since 2024 I have been at Meta Reality Labs, working on machine learning for the neural wristband, a neuromotor interface that decodes surface electromyography at the wrist into input (the system described in Kaifosh et al., Nature, 2025). I work on building novel machine learning capabilities for next-generation devices, and on making the band’s models more robust in production across the various conditions in which the band is used.

Before that I spent over four years at Microsoft on the Industry AI team, focusing on energy and sustainability. I was recruited to continue my doctoral work on deep learning for seismic interpretation and take it to production. That work became the open-source seismic-deeplearning library, which adopted my public facies-classification benchmark as its data source and baseline, and I helped deploy its models for two energy majors. I then worked on LLM-driven data ingestion and methane leak detection from satellite and ground sensors (NeurIPS 2021 CCAI, LOD 2022 Best Paper Special Mention). I later built the analytics engine behind Microsoft Sustainability Manager, in collaboration with Microsoft Research, and took it from first prototype to the general availability of its intelligent insights in 2024. It processed billions of records on Spark for more than 100 enterprise customers, including some of the largest companies in the US, with zero production incidents.

In 2018 and 2019 I was at Airbus Aerial, where I built the team’s infrastructure for training, evaluating, and deploying deep convolutional neural networks, and used it to detect damaged buildings in satellite and drone imagery after hurricanes, wildfires, and other natural disasters. Earlier I interned at Panasonic (2016) on real-time driver monitoring using infrared cameras, and at Mitsubishi Electric Research Laboratories (MERL) (2017) on state-of-the-art facial landmark detection and tracking.

I hold an MS and a PhD in electrical and computer engineering from Georgia Tech. My doctoral research, completed in 2019 under Prof. Ghassan AlRegib, addressed a central limitation of deep learning: its dependence on large annotated datasets. It developed methods for training deep networks when dense annotations are unavailable and expert labeling is prohibitively expensive, combining weakly supervised learning with novel signal representations. The aim was to extend deep learning to domains where annotation is the bottleneck, with 3D seismic interpretation as the primary application. In 2021 I also held an appointment as an assistant professor at King Fahd University of Petroleum and Minerals (KFUPM). I have published 27 journal and conference papers and am a named inventor on 3 published patent applications (publications, Google Scholar).

To reach me, use the form below, with an email address if you want a reply.