Projects
 | DataSynth webapp Devpost Customizable synthetic data generation webapp for training AI models. Developed as part of TreeHacks, Hackathon by Stanford. |
 | Scinetific Document Summarization Summarization of scientific articles is a widely studied problem. Incorporating the additional information like facets of research information along with the summaries of papers written by other papers citing it can be useful for the task of summarization. Citants also contain meta-commentarty over reference paper.CL-SciSumm shared task is a task consisting of summarization of a research paper using its citantations. In task 1, built an improved recall system of reference text spans from a given citing research paper (Task 1A) and constructed better models for comprehension of scientific facets (Task 1B). In task 2, used the reference spans, citances, abstract, etc to construct a summary of the given scientific paper. We experiment on both extractive and abstractive methods and produce summaries on 3 scales, small, medium and large. |
 | Mars Rover webgame Winner all over India in Mars Colonization Program by Microsoft, 2020 Code Worked on Automated mars rover web game. Developed the game in Agent Centric way. Used shortest path-finding algorithms like Collaborative Learning Agents, A*, Dijkstra, Best first search, IDA*, Jump-Point Finders and their bi-directional forms to make the AI rover navigate the mars. Applied Travelling salesmen algorithm and made the AI agent render multiple destinations in the shortest path avoiding all obstacles. Built using Object Oriented programming cocepts. Used Jquery, Rafael.js, and HTML, CSS and javascript. |
 | Smart Library Management System Winner all over India in Smart India Hackathon, 2020 Made Google’s BERT Transformer based auto-tagger via live data extracted using google books api, collaborative filtering recommendation engine; contact-less library: QR based issuing of books, scheduling time-slots with allotted tables; Advance search with voice using Bitap Algorithm; admin dashboard; virtual guide: chatbot(made via dialogflow) which suggests books, option to issue/return, book slots. Used Flutter, Django, Firebase and self hosted AWS server. |
 | Style Transfer Code Change style of images via Deep Neural Network. |
 | Fake News Detection System Code video Developed a web-app using Angular and Django where users can check genuineness of any site by providing it’s URL and get other user’s reviews along with predictions by DL model. Trained various Deep Learning models like LSTM, XGBoost and CNN on three datasets- Kaggle fake news net, Kaggle: getting real about fake news and Kaggle fake news Prediction. Also trained a passive aggressive classifier (online learning algorithm) and incorporated user-rated scraped reviews for real time prediction. |
 | India Statewise Rainfall vs Emissions visualization Code video Developed India State-wise rainfall vs greenhouse gas emissions visualization using Python library Bokeh. |
 | Poisson-Matting Code Did matting for natural images in complex scenes by calculating the gradient of matte from image and solving Poisson equations. Refined matte using semi-supervised approach and alpha-blended to get complex images in new backgrounds |
 | Image Annotator Webapp Code User can select area on image via dragging and label the object; can also upload new images in specific category and annotate it. Deployed in flask. |
 | WikiPedia Search Engine Code Included support for queries ranging from basic to advanced search based on topic, info-box, body, etc and ranked search results from Wikipedia Corpus. Wrote indexer from scratch and indexed entire Wikipedia English corpus |
 | Adversarial Neural Cryptography Code Inspired by Google brain paper "Learning to Protect Communications with Adversarial Neural Cryptography". Work on training neural networks for real time encryption in multi-agent transmission and multiple adversaries. Trained LSTMs Alice and Bob who transmit data protecting it from multiple adversaries trying various adversarial attacks. |
 | Stereo reconstruction and Non-linear optimization Code Generated a dense 3D point cloud reconstruction of a scene from stereo images.Synthesized a new image taken by a virtual monocular camera fixed at any arbitrary position and orientation and recovered the pose using iterative(Perspective-from-n-Points)PnP from scratch. Further imple- mented bundle adjustment and Structure from motion pipelines. |
 | Pose Graph Optimization in SLAM Code Optimized robot’s trajectory using pose graph optimization for 1D and 2D SLAM problems(Simultaneous Localization and Mapping). Implemented in python from scratch as well as using g2o and jax libraries. |
 | GANS from scratch Code Various GANs like pix2pix, DCGAN, StarGAN etc from scratch. |
 | Self Driving Car(in simulation) Code Used Deep Q learning. Passed states of environment vectors as input to neural network. Used three sensors for direction and moment. Built in python from scratch without using libraries. |
 | Reinforcement Learning Algorithms Work on RL algorithms in Robotics and Control under Professor Madhav Krishna. Code Implemented algorithms from Monte-Carlo to PPO, TRPO, DDPG etc from scratch and tried them on various gym evironments. |
 | Genetic Algorithm for optimizing weights of a ML model Code Used weights of model as population, used MSE as fitness function to optimize the performance of model. Used Whole Arithmetic Recombination cross-over and rank based selection. |
 | Digital Image processing Operations Code Various Digital Image processing operations from scratch |
 | Statistical Methods in AI Code Various ML Algorithms from scratch |
 | Wikipedia pages bot Code Used SPARQL and python script to fetch relevant items through Wikidata and converted them to Natural Language. |
 | Artificial Intelligence playing Atari games Doom Code Breakout Code Used Asynchronous Advantage Actor Critic(A3C) algorithm for Breakout and Deep Convolutional Q-Learning with Eligibility Trace for Doom. Built from scratch in python without using libraries. |
 | Trust Reads NewsApp Code Web-app where user can get the news they wants to read on from trusted sources. The news is scraped from trusted sources and displayed to user. Search by script through google search library. Deployed in flask. |
 | Shopping app based on MERN stack Code Functionality to sell and buy products for vendors and users, interactive dashboards, achieves secure login with bcrypt. Automatic dispatch of products when required bulk order is received by vendor. |
 | Jetpack Joyride Terminal Game Code A terminal-based python game, made without any inbuilt game-building library, followed Object Oriented programming paradigm and developed features like dragon, firing, shield, magnet, gravity, coins, etc where all game elements are objects. |
 | Point Cloud Registration and Occupancy grid construction Code |
 | ICP and Non-linear least squares optimization Code |
 | Transformations: Rotations, Eulear angles, gimble lock Code |
 | Chatbot Code Chatbot made using Deep Natural Language Processing. Trained Seq2Seq model. Cornell movie-dialogs corpus Dataset used for Training. |