Projects
This portfolio focuses on embedded AI, edge intelligence, autonomous systems, applied AI assistants, and privacy-preserving machine learning.
Jetson Edge AI Runtime for DonkeyCar
Deploying a camera-LiDAR robotic-control policy to Jetson Nano with ONNX/TensorRT FP16 inference, runtime profiling, asynchronous logging, safety gates, and non-takeover shadow validation.
Focus: Robotics software, edge AI deployment, TensorRT inference, sensor runtime, safety validation, sim-to-real alignment.
Stack: Python / DonkeyCar / PyTorch / ONNX / TensorRT / Jetson / ROS LiDAR
FalCom / APPFL Compressor Acceleration
Python/C ABI integration, native C codec work, OpenMP/SIMD CPU optimization, and guarded CUDA q8 experiments for model-update compression.
Focus: Model-update compression, Python/C ABI integration, CPU SIMD/OpenMP optimization, CUDA feasibility, correctness-gated benchmarking.
Stack: Python / C / CUDA / OpenMP / SIMD / APPFL
Lightweight HAR with Model Distillation
Compressing wearable-sensor activity recognition models through teacher-student distillation for edge devices.
Focus: Time-series classification, model distillation, edge AI, sensor data preprocessing, PAMAP2 / MHEALTH datasets.
Stack: Python / TensorFlow / scikit-learn / NumPy / Pandas
Encrypted Collaborative Filtering
A privacy-preserving recommendation system using encrypted IDs, zero-knowledge proof validation, and matrix factorization.
Focus: Recommendation systems, privacy-preserving ML, AES encryption, zero-knowledge proof, matrix factorization.
Stack: Python / PyTorch / MovieLens / AES / ZKP
Stevens IT Quackathon: Data-Driven Ducks
A Spring 2025 Stevens IT Quackathon team concept for an AI academic and career guidance assistant.
Focus: AI assistant design, academic planning, career guidance, curriculum-aware recommendations, student success workflows.
Stack: Microsoft AI Technology / Recommendation Systems / NLP / Student Success