United States
The evolution of autonomous machines is no longer a challenge of basic AI reasoning, but a challenge of systemic integration at scale. Having spent 19+ years at NVIDIA, I have witnessed and driven the transition from rule-based computer vision to the current era of End-to-End Deep Learning stacks. My focus is on dismantling the traditional silos between Sensing, Prediction, and Planning to achieve a unified, high-performance autonomy architecture. In a world where 99.9% reliability is just the starting point, I lead teams to solve the "Last Mile" of safety-critical AI: ensuring that autonomous delivery, trucking, and passenger vehicles can operate seamlessly in unmapped, high-entropy environments. I believe that true AI excellence is found where hardware-aware software design meets rigorous safety-class validation.
Architecture Leadership: Directed a multi-disciplinary organization of data scientists and ML engineers to redefine the NVIDIA autonomy stack, focusing on the convergence of neural perception and real-time planning modules. Safety-Critical Infrastructure: Engineered the "Offline-to-Online" assurance framework, ensuring that deep learning models comply with ISO 26262 and SOTIF standards before field deployment. Scale & Performance: Optimized large-scale data ingestion and labeling pipelines for petabyte-scale sensor data, reducing the model iteration cycle by 40% through automated corner-case identification. Institutional Result: Established the "Safety-Aware AI Validation" protocol, which became the baseline for all NVIDIA autonomous construction and trucking pilot programs.
Algorithmic Innovation: Developed foundational Deep Learning models for multi-object tracking and lane-level localization under adverse weather conditions. Hardware-Aware ML: Collaborated with GPU architecture teams to optimize CUDA-level inference performance for real-time edge computing on NVIDIA DRIVE platforms. Success Metric: Successfully transitioned three core research prototypes into production-grade features for tier-1 automotive partners.
Core Development: Built the initial generation of data processing tools for early-stage autonomous vehicle testing. Legacy Learning: Directed the migration from classical CV heuristics to early-stage CNNs, identifying the critical need for high-fidelity simulation environments early in the development cycle.
Model Optimization: Focused on model quantization and pruning techniques to enable efficient AI execution on resource-constrained devices.