November '24 - March '25

Matformer - Optimizing Vision Transformers for Long-Tail Image Classification
Guide: Prof. Ganesh Ramakrishnan, IIT Bombay
Conducting research on the Matformer architecture for Vision Transformers, optimizing long-tail image classification and improving Mean Average Precision (MaP) through experimentation with multiple architectural variations

July '24 - May '25

Detection and Mitigation of Hallucination in LegalLLM
Guide: Prof. Pushpak Bhattacharyya, IIT Bombay
Developed a high-precision legal advisory chatbot to provide accurate answers while addressing the issue of AI-generated inaccuracies, known as hallucinations. Used advanced techniques to improve factual reliability, including better information retrieval and response validation, ensuring credible and trustworthy outputs.
Demo Report

May '24 - August '24

Noise Injection in Sequential Models
Guide: Prof. Ganesh Ramakrishnan, IIT Bombay
Researched the effects of noise injection on sequential models like Transformers and LSTMs, developing a modified gradient descent algorithm to improve training stability. This work enhanced model accuracy, robustness, and generalization in NLP tasks, enabling faster convergence and better performance in diverse applications.
Code Report

July '23 - May '24

Recommendation Systems for Smart Homes
Guide: Prof. Pushpak Bhattacharyya, IIT Bombay
We developed a Transformer and Temporal Convolution Network-based model to predict the timing of the next action in smart homes. The model incorporates RBF and Time2Vec embeddings to effectively capture temporal dynamics. By reformulating the regression problem as a classification task, we achieved a 10% improvement in accuracy and significantly reduced RMSE across multiple datasets, enhancing the reliability of smart home automation systems.
Report

July '22 - April '23

Automated Nano-Optical Measurement
Guide: Prof. Anshuman Kumar, IIT Bombay
Python-based automation of optical photoluminescence spectrometric data collection in quantum physics experiments, utilizing communication with a spectrometer and a motorized stage for efficient and time-saving data acquisition.
Code Demo Report

September 2024 - November 2024

JPEG Image Compression
Developed the JPEG image compression algorithm in Python using Discrete Cosine Transform (DCT) and entropy encoding. Conducted a comparative analysis with homogeneous diffusion models to evaluate performance.
Code

May 2023 - July 2023

VanGoghAI
Implemented a Neural Style Transfer system using pre-trained VGG19 to blend real-life images and silhouettes with diverse artistic styles, generating logos and artistic compositions.
Code

July 2023 - September 2023

NLP and Large Language Models
Built advanced NLP systems leveraging transformer architectures and LLMs for sentiment analysis, text classification, and chatbot applications, using techniques like n-grams and transformers.

May 2023 - July 2023

Deep Carlsen
Developed a chess engine inspired by the Giraffe research paper, training it on 300+ features with deep learning and integrating Minimax with Alpha-Beta Pruning for efficient game tree analysis and optimal move selection.
Code

January '23 - April '23

Elevator Control System using FPGA
Designing an elevator control system which relies on hard coding to give us the optimal path of the elevator.
Demo Report

May '22

Python Tic Tac Toe AI Game using PyGame
Developed a GUI Tic-Tac-Toe game which plays with the user using an AI.
Code Demo

January '22 - April '22

Bubble Trouble Game Development
Developed a bubble shooter game using Simplecpp graphics consisting of 10 levels which involve bubbles that bounce off walls, and take a health point from the shooter on hitting him while making him immune to damage for a short time
Code Demo

January '23 - April '23

Analysis of dynamics of regulatory and signaling pathways in the cell
Analysis how signal transmission and response occurs in the cell using feedforward and feedback signals represented by Non Linear Dynamics
Report