My Projects
A showcase of my work and projects that demonstrate my skills and expertise in AI, Deep Learning, and Web Development.
SIH 2025 - Jharkhand Smart Tourism Platform
Developed an eco-tourism platform using Next.js and Flask, featuring a Gemini-powered AI trip planner and AR/VR attraction previews. Integrated a Blockchain-backed trust system for secure bookings and a digital marketplace to empower local artisans. This project achieved Top 50 placement in the SIH 2025 Internal Hackathon, showcasing innovative solutions for sustainable tourism.
Intelligent Skin Condition Diagnosis Using Deep Learning with Explainable AI
Developed a high-precision diagnostic system using MobileNetV2 with transfer learning to classify skin lesions from dermoscopic images, optimized for efficiency on limited hardware. The solution features a robust preprocessing pipeline that employs DullRazor for hair removal and bilateral filtering for noise reduction. Incorporated Explainable AI (GradCAM) to generate heatmaps highlighting irregular borders and utilizes a confidence-based interpretation mechanism to provide actionable recommendations for dermatologists.
Solar PV Detection Pipeline
Built an automated AI-powered pipeline designed to detect and quantify rooftop solar panels using YOLOv8 and satellite imagery. The system manages the end-to-end process of fetching satellite data, performing real-time object detection, and calculating the estimated surface area of installations. The detection logic was specifically optimized to identify panels across various roof types, enabling effective remote energy assessment and data analysis.
Falcon Safety Sentinel — YOLOv8s Object Detection System
Developed an optimized YOLOv8s-based computer vision model to detect critical safety equipment within a digital-twin space environment. The project involved building a complete pipeline—from dataset preparation with synthetic high-fidelity data to training and evaluation—achieving efficient real-time detection. Performance was validated using confusion matrices and precision-recall curves, with advanced hyperparameter tuning and modular scripts implemented to maximize accuracy and inference speed.