Applied Machine Learning Scientist
Building production-grade ML systems at the intersection of autonomy, transportation, and safety.
Ph.D. in Computer Science (Ohio State, 2019). Six-plus years of industry experience at Zoox, Lyft, and Atelio by FIS — designing deep learning systems for autonomous driving perception, large-scale routing optimization, and real-time fraud detection. Advised by Prof. Rajiv Ramnath.
Work
Architected a two-tier ML fraud detection system with a real-time residual learner for continuous adaptation to evolving fraud patterns. Scaled inference pipelines to handle multi-billion daily transactions with low-latency serving infrastructure.
Designed and owned large-scale simulation-based evaluation frameworks for autonomous systems, defining metrics, building auto-rating pipelines, and optimizing data efficiency for complex multi-agent scenarios..
Designed and deployed an ML routing cost model and deep learning ETA prediction system, improving route accuracy by 15% and driver compliance by 10%. Led a 15+ person cross-functional team (ML, SWE, DE, PM) and built end-to-end ML infrastructure serving millions of daily users.
Doctoral research focused on telematics, contextual data analysis, and driving risk prediction. Produced widely-used datasets on traffic accidents and weather events, and published work at top-tier venues including KDD, SIGSPATIAL, and ICWSM.
Research
Open Data
Large-scale datasets published for the research community, collectively achieving over 2,800 upvotes on Kaggle. Topics span traffic accidents, weather events, road construction, congestion, and urban mobility.
A countrywide traffic accident dataset covering 49 US states, collected continuously since February 2016 from traffic APIs, law enforcement agencies, and road-network sensors. Each record includes location, time, weather conditions, and nearby points-of-interest — enabling real-time risk prediction, hotspot analysis, and casualty studies.
View on Kaggle →A countrywide dataset of weather events — including rain, snow, storm, fog, and extreme cold — collected across the United States from 2016 to 2022. Each event is spatiotemporally annotated with severity, duration, and location, making it suitable for weather impact modeling and traffic-weather correlation research.
View on Kaggle →A large-scale dataset of road construction and closure events across the United States, collected from 2016 to 2021. Useful for studying the impact of construction activity on traffic flow, travel time estimation, and route planning under disrupted road conditions.
View on Kaggle →A comprehensive dataset of traffic congestion events across the United States spanning 2016 to 2022. At 14 GB, it is the largest in this collection and is well-suited for large-scale spatiotemporal analysis of traffic flow patterns, congestion propagation, and the influence of external factors on road performance.
View on Kaggle →A dataset of ~14,000 urban trajectories designed for mobility and route behavior research. Covers diverse city environments and is suited for studying pedestrian and vehicle movement patterns, location embedding learning, and urban navigation behavior.
View on Kaggle →