Real‑time Deepfake Detection
An ensemble of EfficientNet and Vision Transformers that detects AI-generated faces and voice deepfakes with 98.7% accuracy. Runs at 30 FPS on edge devices. Used by a fact-checking organization.
Building the future with Deep Learning, Computer Vision, and Scalable Web Systems.
I turn cutting-edge research into real-world applications.
AI researcher, engineer, and problem solver.
I'm a Full-Stack AI Engineer with a passion for pushing the boundaries of machine learning. My work spans from real-time computer vision systems to large language model applications. I've built deepfake detectors, medical image segmentation tools, autonomous drone navigation systems, and intelligent document parsers.
I believe in using AI to solve meaningful problems – from healthcare to disinformation. Every project is an opportunity to learn and contribute to the open-source community.
CNNs, Transformers, GANs
Model deployment, Docker, AWS
PyTorch, TensorFlow, Scikit-learn
OpenCV, YOLOv8, MediaPipe
Modern tools for next‑gen AI.
Real-world applications that push the boundaries of machine learning.
An ensemble of EfficientNet and Vision Transformers that detects AI-generated faces and voice deepfakes with 98.7% accuracy. Runs at 30 FPS on edge devices. Used by a fact-checking organization.
A 3D U‑Net architecture trained on 1000+ CT scans to automatically segment liver tumors. Achieved Dice coefficient of 0.92. Deployed as a HIPAA‑compliant web tool for radiologists.
Vision‑based navigation system for obstacle avoidance and target following using YOLOv8 and depth estimation. Integrated with ROS2 and PX4. Demonstrated in simulated and real forest environments.
A retrieval‑augmented generation pipeline that answers complex queries from 5000+ legal PDFs. Uses Llama 3 8B + FAISS vector store. Reduces legal research time by 70% for a startup.
MediaPipe + LSTM model that translates 100+ American Sign Language gestures into text in real time. Achieves 94% accuracy on live webcam feed. Open‑sourced and used by accessibility groups.
A temporal fusion transformer model that predicts equipment failure 48 hours in advance using IoT sensor data. Deployed on edge gateways, saving a manufacturing client 30% in downtime costs.
Have an ambitious AI project? Reach out – I'm available for consulting, research collaborations, and full‑stack AI development.