Hello, I'm Sourav Kumar 👋

I love Programming and developement in Machine Learning and Deep Learning Technologies. I love Space and reading about Cosmos. I am Python developer and builds stuff using Python. I love to build systems using algorithms and solve complex problems of real world using technology.

Most of the time I work with Data and that means numpy, pandas, scikit-learn, matplotlib and PyTorch. I have also build apps on AWS and apps for Google assistant and ALexa skills.

I also write technical blogs on Medium explaining hard concepts in easy way.

I believe in version control, writing quality code and documenting every project with proper comments and docstrings !

Projects

Retinal Blindness Detection

Diabetic Retinopathy is a disease with an increasing prevalence and the main cause of blindness among working-age population. The risk of severe vision loss can be significantly reduced by timely diagnosis and treatment. I am proposing Deep Learning classification technique using CNN pretrained model resnet152 to classify severity levels of DR ranging from 0 (NO DR) to 4 (Proliferative DR). A GUI based system has been made using Tkinter. Twilio API have been used to Make SMS connectivity to patients possible in case they are not contactable or accesible.

Real time Object Detection

Developed Real time object detection system which detects various types of objects given images by the user in real time. Labels include from various objects to person. This type of system can be very useful in such possible scenarios and can be integrated into the current tech stack : Intrusion detection / Anti-theft surveillance systems, Maintaining safe distance between people during the pandemic times (COVID-19), Detecting face masks on people, Assiting traffic management tasks etc..

Neuralstyletransfer-showcase

This was a volunteering project aims to acheive neural style transfer using the algorithm given by Paper Image Style Transfer Using Convolutional Neural Networks by Gatys, implemented in Pytorch. minimized the content losses and style losses to get a good mix of visuals with the final loss of around 100K. Designed AI generated images which were printed on Tees and sales produced through it was given for charity.

Federatedencryption-showcase

This project aims to achieve federated learning training of model and then testing on encrypted dataset and model using pytorch and pysyft. Federated Learning is a collaborative form of machine learning where the training process is distributed among many users and not based totally on a single server. Here , the server is the main control for coordinating everything but most of the work for training process is done by federation of users.

NeuralNets-Pure-Python

Objective : This is a mini tutorial which aims to develop intuition about how matrices gets/changes/modifies their shapes as they go from layer to layer in a neural network. we are building more complex neural networks becasue layers stacked over one another keeps the matrices computations overly abstracted but getting deeper insight into their shapes will help us in understanding how our inputs and outputs are related.

MicrosoftBing-search-query-prediction

The world is going through COVID-19 pandemic caused due to Novel Coronavirus and people have mostly stayed home and searched various things related to Coronavirus on internet using popular search engines such as Chrome, Bing, Yahoo, DuckDuckGo etc. I have explored the intent of the people by just using the searches and also build a predicting model predicting the Country of origin from where the search query was issued. This analysis will really help central authorities to take appropriate action and eventually helping out and benefitting the Citizens.