A system to visualize custom tiled floor in a room. Taking a indoor image and the tile image as input and returning the indoor image with the custom tiled floor while retaining the shadow and lighting information.

A system with sensor fusion and supporting neural networks to autonomously drive a large golf cart like vehicle. LIDAR, Stereo Camera, GPS are used for various different tasks. The entire system is deployed on an edge device and autonomously driven around the campus.

A system to predict footballing actions from a video. Implemented and trained a branched LSTM architecture for the prediction, building data preparation and ingestion pipeline for training and inference.

A system to predict signal attenuation based on temporal and visual information. Implemented a branched architecture with 3D Convolutions and LSTMs with priors and skip connections.

Data Science competition organized by the Israeli Ministry of Defence. It involved classification of human and animal classes based on RADAR signals. Achieved 23rd Rank globally. Only 25 out of 300 teams were able to achieve AUC score above 0.90.

Unofficial implementation of paper by Google Research - Representation Learning for Information Extraction from Form-like Documents. Based on transformers to detect entities in an invoice or receipts. Implemented using Pytorch.

Video Analytics system to detect social distancing and face-mask violations on CCTV cameras. Deployed using Deepstream 5.0 with TensorRT optimized neural networks. Runs on 20 live stream feeds simultaneously and notifies violations to the admin.

Open-source full-featured ICU ventilator designed with rapid manufacturability in mind. Delivering treatment options including high flow oxygen therapy and mechanical ventilation to communities that have limited or no access to these life-saving interventions during and beyond the COVID-19 pandemic.

A comprehensive document intelligence system (Akshar) which uses AI, NLP and OCR to ingest and extract details from variety of documents including pdfs, images, document format.

A service for extracting text from ID cards in India, like Aadhar Card, PAN Card and Driving Licence. You just need to click and send a picture of the card to the API and get a json with your details. It uses Connectionist Text Proposal Network (Open Source) along with Tesseract for text extraction.

A deep learning based system for disorder detection in tomato plants. Also used IoT to get sensor data from the plants. Two seperate models were trained for this project. One dealt with Images and other with Sensor Data. You can use both to predict the Disorder of Tomato Plants.

A web app to convert handwritten forms to digital forms. Initially you are supposed to upload a template of your form that isn't filled. That is only a one time task. If the templates of your form is already available, you just need to upload your handwritten form and it will be converted to digital text.

A service for people to report the potholes in their area. It is a telegram service bot wherein you can report the image and location of the pothole. The complaint will be lodged. The admin also gets the severity of the pothole and can act accordingly.

An Android application that can be used to check-in registered participants using the QR Code that was sent after successfull registration. It was specially built to use during ERR_404 State Level Hackathon. This app can help avoid long queues and organize an event better.

A barebone model built for detecting weapons in an image. It was built using YOLOv2 (You Only Look Once Algo version2). The model is built upon Darknet YOLO but is also ported on Tensorflow Lite, Protobuf file and also in YAD2K (Keras port of YOLO).

This project was developed during 24hr Hackathon - Unscript 2k19. It is a service as telegram bot that takes infected skin image as input and predicts the skin disease. It was trained on Dermnet Dataset for Skin Diseases.

Designed and Developed a friend's Portfolio Website. Built on Dark Theme. What is unique about it is, his profile being that of an hacker, it comprises of a terminal. It displays details about him from that terminal. Tried to develop it such that it resembles his profile of an hacker.