Research and engineering projects where the interesting part was always the problem behind the decisions — and the tradeoffs that came with them.
The ones with the most interesting problems and the biggest impact.
ML Framework
TrainConv Framework
A distributed training framework on PyTorch and TensorFlow with configuration-based pipelines and automated orchestration.
- +Config-driven training pipelines
- +Tune hyperparameter search
- +TensorBoard monitoring
2022Research · IEEE AISP
GCN for Pandemic Prediction
Graph convolutional networks for predicting state-wise COVID-19 incidence with up to 85.3% accuracy.
- +85.3% prediction accuracy
- +Graph convolutional networks
- +Published at IEEE AISP 2022
More work, from computer vision to embedded systems and simulation.
Computer Vision
Real-time Ship Detection
A modified YOLOV3 architecture for ship detection in synthetic radar images with high accuracy and performance.
- +81.02% mAP on SAR imagery
- +30+ FPS real-time
- +Modified YOLOv3 + attention
Embedded · Healthcare
Programmable Insulin Delivery System
Raspberry Pi-based closed-loop system for automated insulin delivery, leveraging continuous glucose monitoring and real-time processing.
- +Closed-loop CGM control
- +Raspberry Pi + micropump
- +18-month funded project
2021Simulation · Research
CaprioSim
Traffic Sign Recognition and Flow Simulation using Cellular Automata. Integrates computer vision-based traffic sign detection with cellular automata simulation to model priority-based traffic flow at intersections.
- +Traffic sign recognition
- +Cellular automata flow model
- +Presented at SSNCE 2021