Projects

I build data-driven projects using modern machine learning, geospatial analytics, and visualization tools, with a strong focus on accuracy, stability, and clear evaluation. I pay attention to clean preprocessing, reproducible workflows, and meaningful metrics so results are reliable and easy to interpret. I’m actively expanding this portfolio and will continue adding more end-to-end projects and case studies over time.

REFocus: Real-World Motion Blur Recovery & Optional Colorization (DeblurGAN + U-Net/CycleGAN)

Built a two-stage, luminance-first pipeline that deblurs real-camera images on L (Lab) with DeblurGAN and conditionally colorizes grayscale inputs with U-Net & CycleGAN, delivered with a web interface.

Code Demo Poster

Climate and Disaster Prediction Recommender System

This project aims to develop a recommender system that predicts climate temperature trends and assesses potential natural disasters. The system provides valuable insights into environmental risks by leveraging historical data and predictive modeling.

Code Demo Presentation

Cows_and_Bulls_Game

Cows and Bulls is a web-based classic number guessing game reimagined with concepts from information theory, such as entropy and mutual information, to make intelligent and efficient guesses.

Code Demo Presentation

VisionSaver

A comprehensive system to detect ocular diseases through retinal images using Machine Learning (ML) and Deep Learning (DL) algorithms like ResNet, and Random Forest using fundus photographs aiding healthcare professionals.

Code Demo Paper

Suspicious-Activity-Detection-using-LRCN

This is a intelligent surveillance system uses an LRCN model to detect suspicious activity in CCTV footage. Developed with Python, Keras, and TensorFlow, automating by identifying abnormal human behaviors through deep learning and visual pattern recognition.

Code Paper

Handwritten recognition using CNN

Published at an international IEEE conference, the system recognizes handwritten text and converts it into editable text in multiple languages. It utilizes Convolutional Neural Networks (CNNs) for recognition and Natural Language Processing (NLP) for efficient data handling.

Code Paper

Improving Ads Productivity by 50%

Developed a deep learning model using LSTM to predict ad extensions from website links. Data is gathered via web scraping, preprocessed with NLP and feature encoding, and integrated with a user-friendly interface for accessibility.

Code Demo Paper

AI-based Chatbot

Developed a multilingual chatbot using the Rasa framework, Python, and Web Speech API to resolve AICTE-related queries. The project integrates HTML, CSS, JavaScript for the frontend and REST API with Socket.IO for seamless connectivity.

Code Demo Poster