Work
Professional experience — roles and impact.
How I work
I build real-time ML and robotics systems end-to-end — from sensor fusion and tracking to embedded deployment. I care about measurable performance (latency/robustness), clean interfaces, and reproducible experiments. I profile early, validate with ground truth where possible, and iterate based on failure modes.
Advanced Research Engineer (Co-op) · Toyota Material Handling (TMHNA)
Greene, NY
May 2025 – May 2026
Event + RGB fusion perceptionReal-time multi-camera bird’s-eye view (BEV) on embeded GPUOperator-assist autonomy prototypes (patent pending)ROS2 + C++ real-time robotics integrationCAN-based control + safety arbitration layer
- Researched event-based cameras for pedestrian/obstacle perception and benchmarked them against RGB baselines; built an event+RGB fusion prototype to improve detection in high-dynamic-range scenes where standard cameras saturate.
- Engineered a real-time bird’s-eye-view (BEV) compositor on embedded hardware: performed multi-camera calibration, optimized the frame pipeline with a zero-copy strategy, and implemented the runtime in C++ to maintain low-latency visualization on Jetson Orin Nano.
- Developed a suite of operator-assist capabilities that combine perception + kinematics to generate real-time guidance overlays and repeatable maneuvers; contributed to a patent filing (pending) related to perception-driven guidance for industrial autonomy.
Research Assistant · Rochester Institute of Technology (RIT)
Rochester, NY
January 2025 – May 2025
SINDy for discovering ODEs from simulation dataFilament growth with severing dynamicsUnlimited vs limited monomer pool regimesModel validation via steady-state + rollout checks
- Built a SINDy pipeline to learn compact differential equations for filament length dynamics from stochastic simulation time-series.
- Implemented preprocessing for noisy signals (averaging, smoothing, numerical differentiation) so SINDy could fit stable, interpretable models.
- Validated discovered dynamics by comparing predicted vs simulated dL/dt and checking convergence to steady-state length under fixed parameters.