Built a cloud-based, end-to-end solution for manipulating INSAT satellite data, tailored for meteorologists and scientists at ISRO. Contributed to the data processing pipeline by projecting Level 1 INSAT-3DR data and converting it into cloud-optimized GeoTIFF formats. Implemented download functionality and API playground to stream data directly into notebooks. Won Smart India Hackathon 2024 for Problem Statement SIH1738: "Innovative applications of cloud-optimized GeoTIFFs for INSAT satellite data." Received guidance directly from ISRO scientists and had the opportunity to meet Nilesh M. Desai, Director of SAC ISRO.

Built a Convolutional Neural Network (CNN) using PyTorch to classify different rice varieties based on images. The model was trained on a dataset of rice images split into training, validation, and test sets. Implemented data preprocessing, model training, and evaluation workflows with visualization of training metrics. Published a detailed tutorial on Towards AI explaining the step-by-step process of building and training the model.

Developed an AI-powered personal journaling assistant using the Conversational Realtime GPT API to address the challenges of traditional journaling. The platform facilitates voice-based journaling, analyzes emotions, tracks habits, and generates data-driven insights on personal well-being. Created an intuitive interface for users to maintain consistency in documenting their thoughts while providing quantifiable insights into mood and behavioral patterns over time.

This project addresses a real-world challenge by implementing a binary classification model to determine whether a chicken is healthy or suffering from Coccidiosis based on fecal images. Developed a deep learning model using VGG16 CNN with transfer learning. Implemented Data Version Control (DVC) for managing datasets, models, and experiment tracking. Deployed the application on Azure Container Registry and hosted it via containerized infrastructure.

Developed a local AI chatbot with a custom personality - a chess-loving little girl named Emily. Built using Ollama's Phi 3 model that runs locally without requiring an API key or GPU. Implemented speech-to-text and text-to-speech capabilities for natural conversation flow. Created a comprehensive tutorial explaining how to set up Ollama, customize AI personalities, and build voice-interactive chatbots with Python.

Created a web application that helps parents generate artistic QR codes containing their child's information that can be attached to their bags or clothing. When scanned, these QR codes provide the child's contact details to help reunite lost children with their families. Used generative AI to create visually appealing QR codes that blend into designs while remaining scannable. Implemented with Streamlit for the frontend UI and Azure Blob Storage for hosting generated PDF files containing child information. The project was published in the International Journal of Intelligent Systems and Applications in Engineering.