DEEP LEARNING • ML SYSTEMS • PROBABILISTIC MODELING
Building AI systems from research to real-world deployment.
Foundational research for Infima.io & WhiteRabbit.ai
Across ML, healthcare, finance, and applied modeling
ML models deployed in Amazon robots across US
I am a researcher and engineer working at the intersection of deep learning, probabilistic modeling, and ML systems. I also served as an Adjunct Lecturer at Stanford University. My work combines modern AI with optimization, stochastic modeling, and statistics to solve real-world problems across robotics, healthcare, finance, and operations. My research at Stanford formed the technical foundations of two AI startups, Infima.io and WhiteRabbit.ai. I spent five years at Google Brain developing deep learning systems and research, and now lead ML research for robotics at Amazon Lab126.
Research featured in: HPCwire · SiliconANGLE · AWS · PCMag · Stanford GSB · CNBC · MIT Tech Review
10+ years at Google Brain and Amazon Lab126, alongside academia at Stanford.
Adjunct Lecturer and graduate research mentor at Stanford University.
Revamped curriculum and served as Principal Instructor for 3 years, teaching graduate students across Stanford and industry professionals remotely via SCPD.
Mentoring PhD students and industry researchers across ML and applied mathematics.
Recent research and earlier work with lasting real-world impact.
Multimodal learning · 3D vision · robotics · sequence modeling
Improves motion–language alignment using LLM-based text canonicalization to reduce noisy caption supervision before contrastive learning, delivering stronger cross-dataset transfer and state-of-the-art motion–text retrieval.
Uses monocular depth cues from foundation models to improve 2D-to-3D human pose estimation without additional sensors, improving cross-dataset generalization across multiple pose architectures and benchmarks.
Learns socially appropriate robot parking locations from human preferences, aligning a robot’s objective with how people actually want it to behave in the home. Evaluated using preferences from people who lived with a social robot.
Combines permutation-invariant set representations with modern sequence models to learn temporal and cross-sectional structure across collections of time series, with applications to mortgage-risk prediction and portfolio optimization.
Foundational research behind two AI startups
Applied deep learning at massive scale to model nonlinear mortgage dynamics across 120M+ U.S. loans. The research helped form the technical foundation of Infima, later acquired by AD&Co.
Developed deep-learning computer vision models for automated grading of diabetic retinopathy from high-resolution retinal images. The research helped form the technical foundation of WhiteRabbit.ai.
