My Projects

An OCR + NLP pipeline using MISHA, OpenCV, PyTorch, and BERT to recover faded fax text and predict unrecovered text.

Benchmarked relational (PostgreSQL) vs document (MongoDB) data models, achieving 90% faster queries via optimized indexing and pre-joined collections.

A self-supervised model for feature retrieval combining DenseNet with Barlow Twins loss for image similarity tasks.

Improved BERT accuracy & stability via a new AdaptiveAdam optimizer with Gradient Centralization & Clipping.

A deep learning based web application that uses PoseNet to detect Yoga Postures. Worked with in-browser libraries for machine learning like p5.js and ml5.js and built a web application in Node.js runtime that classifies Sun Salutation Poses using a neural network.

Created 2 standard chess datasets which can be used by beginners in computer vision field. Applied Image Manipulation Techniques (using PIL module) and attempted classification algorithms like SVM, KNN, CNN.

A MEAN Stack application that connects youth who have a passion to teach with schools that are in need of teachers for various skills/areas. Worked with Node.js run time environment and wrote queries on a no-sql database through Express.js framework. Also, implemented the integration of various APIs for email-sending, text message sending and video call hosting features.

Re-implementations of csvkit in Haskell and Rust. Used various modules in Haskell to build tools like csvgrep, csvstack, csvsql, csvlook from the csvkit toolset.

A collection of various small data science projects and exercises.