Machine Learning & Systems Engineer | Edge AI, Sensor Fusion, Signal Processing & Robust Deployment
I am an ML Engineer focused on deploying high-performance deep learning models to resource-constrained edge devices. My engineering philosophy is rooted in systems reliability: before specializing in AI, I spent two years designing SIL4 safety-critical Train Control Systems at Alstom. I don't just optimize for accuracy in a notebook; I optimize for fail-safe execution, deterministic latency, and real-world edge cases.
Currently wrapping up my MS in Electrical and Computer Engineering at Northeastern University and recently engineered real-time radar/camera fusion pipelines at Aura Intelligent Systems.
I am actively seeking full-time roles where I can build reliable, deployment-ready AI systems.
- Core Machine Learning: PyTorch, Deep Learning (CNNs/ViTs), Time-Series Forecasting, Signal Processing (DSP)
- Edge AI & C++ Deployment: C++, Python, ONNX Runtime, TensorRT, CUDA, TFLite, Model Compression & Pruning
- Computer Vision & Sensor Fusion: OpenCV, YOLO, Point Cloud Library (PCL), Radar/Camera Fusion, ROS/ROS2
- MLOps & Infrastructure: GCP (Vertex AI), Docker, Apache Airflow, MLflow, CI/CD (GitHub Actions), FastAPI, CMake
- Robust Systems Engineering: SIL4 Safety-Critical Architecture, Hardware-in-the-Loop (HIL), Adversarial Robustness
Engineered a custom C++ inference engine from scratch to bypass Python's GIL for real-time robotic perception.
- Impact: Increased throughput by 1.36x (15.0 FPS to 45.0 FPS) on CPU-only hardware and implemented zero-copy memory management.
- Tech Stack:
C++,ONNX Runtime,OpenCV,CMake,YOLO11 - View Repository ->
Designed a real-time causal streaming pipeline to classify high-density ECoG sensor data for brain-computer interfaces.
- Impact: Won 2nd place by compressing the model size to 150KB (16x reduction) to fit wearable memory limits, while hitting a strict 35ms latency budget.
- Tech Stack:
PyTorch,Signal Processing (FFT),TorchScript JIT,Model Compression - View Repository ->
Architected an automated ML pipeline on the cloud for time-series anomaly detection and forecasting.
- Impact: Reduced manual retraining by 80% using statistical data drift triggers (K-S test) and accelerated release cycles by 40% via containerized CI/CD workflows.
- Tech Stack:
GCP (Vertex AI),Apache Airflow,Docker,FastAPI,MLflow - View Repository ->
Investigated edge-case failures and safety vulnerabilities in deploying deep learning models for high-reliability systems.
- Impact: Engineered a hybrid defense strategy recovering >75% robustness under rigorous PGD sensor attacks, proving Vision Transformers are 11% more resilient to localized noise than CNNs.
- Tech Stack:
Vision Transformers (ViT),CNNs,Adversarial ML (PGD/Patch) - View Repository ->