This is a very basic Flask based REST server which takes in the stress-ng factors like compute, storage and network along with time and allows to run time-varying stress tests. stress-ng is an existing CLI tool. This project acts as a wrapper around it.

This project is a sign language alphabet recognizer using Python, OpenCV and TensorFlow for training the InceptionV3 model, a convolutional neural network model for classification.

Built a deep neural network that functions as part of end-to-end automatic speech recognition (ASR) pipeline. I used LibreSpeech dataset to train and evaluate models. This algorithm first converts any raw audio to feature representation that are commonly used for ASR, and then another network maps these audio features to transcribed text.

This project is my introductory project to the world of Deep Learning. I trained an image classifier to recognize different species of flowers. I used Maria-Elena Nilsback and Andrew Zisserman’s dataset of 102 flower categories(http://www.robots.ox.ac.uk/~vgg/data/flowers/102/index.html) for training. The flowers are chosen to be flower commonly occurring in the United Kingdom. Each class consists of between 40 and 258 images. I used ResNet as my pre-trained model for feature extraction and the following structure is of the classifier model layers on top of it

Using calibrated electronic sensors and Arduino board, I collected gaseous data and olfactory chemical substrates of various fruits. The fruit of every type was further in 3 categories of under-ripe, ripe and overripe. The overall outcome was a portable electronic nose like device which would predict the state of the fruit. The multiclass classification was used to achieve this task.

Created a Google Glasses like AR device to assist Autistic kids to understand emotions and help them with constant learning and recitation.

Developed a Flutter based cross platform for analysing and presenting the data which was given by CiSCO’s Meraki Camera. Along with that, provided future prediction of queue analysis by implementing LSTM. Entire features and outcomes were also conveyed by Google Assistant and Alexa, along with automated messaging services in local language support for non-feature mobile phones. This ecosystem had provision for Admin warning in the form of automated SMS as well as generated mail.

An all in one device for visually impaired person. Scene description, OCR, navigation, path detection and Integrated Payments system. Targeted at low income sections.

A portable Braille for visually impaired people. Low cost effective mechanical-electronic device.

Developed MySQL Database managing publications of TIET. In addition to that made the website accessing the Database. The Server-End of the Website was based on PHP.