PredAID: Cancer screening [http://predaid.blockappsai.com/]
* Used angular best practices for better application development.
* Improved app initial load time from 5 seconds to 1 second.
* Developed multi-level real-time notification and role-based access control.
* Used RxJS and NgRx for better API handling and state management.
* Used source-map-explorer to reduce app size by 40%.
* Built RESTful APIs and databases using Flask and MongoDB for PredAID backend.
* The application was selected for demo at multiple hospitals in Bangalore and Kerala, India.
Safe Kerala: Traffic violation detection
* Trained and evaluated a deep learning model to achieve MAP of 96.5.
* Improved model FPS from 6 images/second to 12 images/second by building a custom tracker.
* Used Tensorrt and Deepstream to deploy models on low-powered NVIDIA Jetson Nano.
* The application was selected by the Kerala government to detect traffic violations in its cities.
Testify: Automated Driver License Test
* Used pure OpenCV to measure speed, motion orientation and detect whether the vehicle crossed the white border to identify whether the person passed the driving test.