Hi! I am Aditya Thakur

Designed an image recognition model based on ResNet architecture for low memory devices with the trained model of size approx. 20MB.
Used a combination on regularization and normalization techniques like RandomCrop, BatchNorm, Dropout, Gradient Clipping and Squeeze & Excitation to effectively reduce test loss and generalize the model. With 5M learn-able parameters, achieved an accuracy of 96% on validation set with 200 epochs of training on CIFAR-10 dataset.

• GitHub

Implemented Principal component analysis with MATLAB on a teapots dataset to improve the images.
Principal component analysis is used to reduce the number of variables of a data set, while preserving as much information as possible.

• GitHub

Built a Client - Server model based chat application using socket programming and multi-threading in Python.
The application uses tkinter for basic GUI and socket programming to implement client-server model.

• GitHub

Notebooks explaining tools with examples to tune performance of python programs
Tools covered: Timeit, Decorators, Loops optimization, Itertools Combinatoric iterators

• GitHub

Designed an accident reporting system using Neural Network for classifying frames, achieving anaccuracy of 85%.
Developed a web app using the Flask framework for calling the classifier API and sending e-mail notifications.

Led the development of electrical subsystem comprising of controllers, battery, motor, sensors and other peripherals.
Worked on interactive dashboard & enabling real-time data sharing of vehicle statistics and location.
Implemented a system to detect driver fatigue and microsleep using eye-feature extraction

Designed a system to preprocess and enhance the signal before feeding it to a microcontroller.
Created extensive training data to increase the predictability of gestures. Achieved accuracy close to 70%.