Experience

LiDAR segmentation · Sensor fusion · Spatial analytics

• Developed and deployed a LiDAR semantic segmentation pipeline using Point Transformer v2, reaching 78.3% mean IoU across 11 highway infrastructure classes.
• Designed a sensor fusion pipeline that combines YOLOv8 vehicle detection, ByteTrack multi-object tracking, and GPS/IMU data for mobile curbside surveys.
• Automated acquisition and structured parsing of 400+ North American traffic accident reports to generate statistical summaries of crash types and contributing factors.
• Analyzed intersection crash data to evaluate the policy impact of speed-enforcement changes and documented findings for transportation stakeholders.

Clinical ML · Survival analysis · Feature engineering

• Processed 58K+ clinical data points from 934 pediatric Crohn's patients using mRMR feature selection, reducing dimensionality by ~80%.
• Trained and tuned survival models (Random Survival Forest, Cox PH) achieving a 96% C-index while optimizing training with multithreading (~40% faster).
• Co-authored a manuscript on machine learning prediction of first bowel resection in pediatric Crohn's disease patients.

Data pipelines · Infrastructure · Geospatial tooling

• Re-architected a Python data pipeline with concurrency and targeted profiling, achieving a 5.5× speedup on core jobs.
• Deployed a $6K NAS solution to replace cloud storage, saving approximately $94K annually while improving data locality.
• Integrated ArcGIS Pro workflows with automated validation across heterogeneous geospatial data sources.

Object tracking · Remote sensing · Visualization

• Migrated a vehicle tracking system from SORT to DeepSORT, improving identity persistence and overall tracking reliability.
• Built a wildfire visualization web app with an automated satellite imagery pipeline for monitoring and analysis.
• Ported 400+ lines of MATLAB processing code to Python, reducing friction for future research and tooling.