I am passionate about using data-driven and physics based modeling to building products at scale. I spent the first decade after my undergraduate doing research in academia and the next twelve years bringing innovative and state-of-the-art research into product. The following summarizes some of my projects to date:

• Computer Vision: I built semi-supervised learning algorithms to aid blind users of Facebook effectively interact with images. This was done by devising algorithms for automated detection of thousands of concepts at scale that was embedded in the Alternative Alt Text (AAT) interface. This was covered by several media outlets [forbes, engadget].
• Continual Learning: I was the Tech Lead for the continual learning project in Facebook AI where we built the state of the art continuous learning techniques by developing/adapting SoTA techniques including multimodal self-supervision, memory rehearsal, prompt-based, and backwards compatible platform for continuous learning. This was a horizontal technology that showed significant performance boost across diverse use-cases including integrity violation detection, media and creator ranking in Instagram, covid/vaccine misinformation, and election misinformation, with 3-10x improvement in recall at the operating precisions. More details are provided here.
• Medical Imaging: I developed algorithms for automated detection of stenoses (vessel narrowing) from CT scans across a diverse set of lesion morphologies that became the State of the Art technique for assessment of vessel stenosis. This is now used routinely in the hospital to diagnose patients with heart disease and was published at MICCAI, the premier medical imaging conference. In addition, I also developed algorithms for automated detection of image quality. Read more here.
• AI in Computational Medicine: I developed AI algorithms for simulation of blood flow and pressure in the human heart. These achieved state-of-the-art performance and was cleared by the FDA to be used in the clinic. Read more about the technique here and here.
• Computational Neuroscience: I developed automated quantification of neuronal images from behaving animals and pipelines for graph-based analysis of these neurons which led to the discovery of the presence of hub neurons. This resulted in a co-first author paper published in Nature. Please read here for more details.