My Projects
A showcase of production-ready applications demonstrating expertise in AI/ML, computer vision, and full-stack development.
Transform hand-drawn sketches into photorealistic 2D images and 3D models. Implemented advanced diffusion models (LCTN + Stable Diffusion) for high-quality image generation. Fine-tuned Hunyuan3D-2 for 3D mesh generation. Full-stack application with interactive 3D viewer.
Technical Details
- Sketch-to-Image: Custom LCTN model + Stable Diffusion (768x768 resolution)
- Image-to-3D: Hunyuan3D-2 with configurable parameters
- Interactive Canvas: Fabric.js with brush, shapes, layers, undo/redo
- 3D Viewer: Three.js with real-time rendering and GLB export
- Backend: FastAPI with JWT authentication, image processing pipeline
- Frontend: Next.js 15 with Tailwind CSS, Framer Motion animations
- Database: MongoDB for user data and asset management
Potato Disease Detection
Real-Time Classification with 95.2% Accuracy
July 2025 - Aug 2025
Real-time potato disease classification system helping farmers identify crop diseases early. Enhanced 3+ CNN models for disease detection. Built end-to-end ML pipeline from preprocessing to deployment. Created interactive web interface with analytics and visualizations.
Technical Details
- Model Accuracy: 95.2% on test set
- CNN Architectures: Multiple models tested and optimized
- Data Pipeline: Preprocessing, augmentation, training, evaluation
- Deployment: FastAPI backend with real-time inference
- Frontend: Interactive web interface with analytics
- Visualization: Chart.js for performance metrics and disease insights
Multi-Model CNN Ensemble
12-Class Animal Species Classification
Sophisticated multi-model CNN ensemble using feature-level fusion. Implements 3 specialized models for different animal groups, extracts penultimate layer features, and fuses them for unified 12-class prediction. Demonstrates advanced ML concepts including ensemble learning and feature engineering.
Technical Details
- 3 Specialized Models: 5-class, 5-class, 2-class specialists
- Feature-Level Fusion: 640-dimensional concatenated feature vector
- Architectures: Separable convolutions, batch normalization, dropout regularization
- Fusion Classifier: Dense layers with L2 regularization
- 12 Animal Classes: Across canines, primates, African wildlife
Real-Time Age & Gender Detection
Facial Analysis System
July 2024 - Aug 2024
Real-time age and gender detection system for images and videos. Implements face detection and prediction pipelines with web-based interface for easy access and testing.
Technical Details
- Real-time face detection with OpenCV
- Age and gender prediction pipelines
- Web-based interface for testing
- Support for both images and video streams