Projects

Investigating the creativity safety tradeoff in RL for language models

End-to-end RAG system for factual Question Answering about Pittsburgh and Carnegie Mellon University

Ranking a series of burst images based on aesthetic and technical image quality using deep learning, to find or create the perfect photo.

Implemented a robust method through which we can accurately apply PCA to high dimensional data having outliers.

Implemented reinforcement learning for the game of Connect4 with players using the Monte Carlo Tree Search algorithm and the Q-Learning algorithm.

Prithvi is a Google Chrome extension that notifies the user about greener alternatives and  provides the best eco-friendly results for all product searches. 

Implemented a path planning algorithm called Rapidly-exploring Random Trees (RRT and RRT*) for the navigation of a robot from start to a goal region in a given bounded area with obstacles.

A 3 layer neural network from scratch to identify digits from the MNIST dataset of thousands of handwritten images with an accuracy of 90% on Kaggle. 

Given a propositional logic formula in 3-CNF form, the proposed model maximizes the percentage of satisfied clauses in the formula, using a variation of the genetic algorithm.