Experience
My experience spans academic research and five years of production software engineering before the PhD. That industry background is part of how I work: I build research code that is meant to run, not only to demonstrate.
- Designed and built an experimental platform for driver-bicyclist interactions using integrated RGB camera and LiDAR sensing with computer vision pipelines; 98.1% overtaking-detection recall with a single false positive on 315 real-world events.
- Built a fine-grained vehicle body-type classifier by fine-tuning a vision transformer on a custom injury-risk dataset I assembled and labeled; 89% accuracy on held-out sites with no retraining, with confidence-based abstention.
- Designed a projective-geometry lateral passing-distance estimator from bounding-box features (0.126 m MAE) that needs no camera calibration, and showed the feature design transfers to the Waymo Open Dataset (R² = 0.812).
- Led manuscript preparation across multiple venues; mentored undergraduate and master's students in computer vision, Python data pipelines, and experimental design, including supervision of open-source tooling.
- Led cross-functional development teams and shipped production software across commercial, healthcare, and education clients.
- Full-stack development and UI technology work; delivered internal technical trainings and mentored junior engineers.
- Shipped enterprise features for developer-tools products; recognized with a company hackathon award (2nd place).
- Built university management systems (exam cell, leave management) across the full development lifecycle, and taught undergraduate AI and mathematics courses.
Technical Skills
Programming: Python (TensorFlow, PyTorch, scikit-learn, OpenCV, pandas), Git, LaTeX; working knowledge of Julia, R, MATLAB, C#, SQL. HPC: UM Great Lakes cluster (Slurm).
Computer vision & ML: Geometry-informed computer vision, object detection and tracking (RT-DETR, ByteTrack, YOLOv5, ViT), physics-grounded measurement, out-of-distribution robustness, confidence-threshold abstention, camera + LiDAR sensor fusion, deep learning (CNNs, RNNs, transformers), ensemble methods, time series.
Practices: Open-source release, reproducible research workflows, experimental design, cross-validation, statistical analysis, technical writing, IRB protocols.
Familiar (self-taught, not yet used in a published project): vision-language models, ROS.