Featured Projects
Innovative AI solutions demonstrating technical excellence and real-world impact
Building a Generative Diffusion Engine from Scratch
Built, debugged, and trained a Denoising Diffusion Probabilistic Model (DDPM) entirely from scratch using raw PyTorch. Features a custom YOLO-style U-Net with Sinusoidal Position Embeddings and a mathematically rigorous linear beta schedule. Scaled to train on the full 60,000-image MNIST dataset.
.png)
.png)
.png)
.png)
.png)
.png)
.png)
.png)
.png)
Traffic Monitoring System
Advanced real-time vehicle tracking system using YOLOv8 object detection combined with SORT algorithm for multi-object tracking. Integrated EasyOCR for automatic license plate extraction and recognition with 90%+ accuracy in varied lighting conditions.
RAG Model Framework
Comprehensive CLI and Flask-based application for intelligent document Q&A. Supports multiple formats including PDF, DOCX, and XLSX. Leverages state-of-the-art Transformer models for accurate information retrieval and contextual understanding.
Test Cheating Detection
AI-powered exam integrity monitoring system utilizing YOLO for real-time object detection and MediaPipe for precise head-pose estimation. Detects unauthorized devices and suspicious behaviors with automated alert generation for proctors.
YOLO Object Detection Suite
Focuses on object detection using YOLOv11 and other YOLO models. Automates identifying objects in images, applicable for surveillance, quality control, and smart monitoring.
Multi-Class Object & Text Recognition
A comprehensive computer vision pipeline that detects humans and animals in images and videos, classifies them correctly, and simultaneously performs Optical Character Recognition (OCR) on the media.
Real-Time Drone Tracking System
Advanced computer vision system to accurately detect and track drones in varied environments using YOLO and SSD (Single Shot MultiBox Detector) architectures.
Robust QR Code Detection
A robust QR code detection and decoding pipeline using OpenCV. Designed to handle challenging real-world conditions including noisy, rotated, or low-contrast inputs.
Drone Image Classification Research
A research-oriented comparative analysis of two prominent deep learning approaches (CNN and ResNet50) for highly accurate drone image classification tasks.
