- ✉️ You can contact me at abdelrahmantawfik740@gmail.com
- 🧠 I'm learning Computer Vision and Generative AI (LLMs & RAG systems)
- 🤝 I'm open to collaborating on AI/ML projects and research
- 📚 I am currently reading Natural Language Processing with Transformers and Hands-On LLMs
- ⚡ Also, have experience in MLOps and deploying ML models to production
| Surveillance System | Pizza Store Hygiene Monitoring System |
|
|
|
| 🏆 1st Place Winner - Built an advanced computer vision pipeline for retail environment monitoring, processing 120+ individuals with 96%+ recognition accuracy. Combined state-of-the-art models (YOLOv11, BoostTrack++, InsightFace) with optimized FAISS indexing for real-time analytics. Fine-tuned on 16GB custom dataset and achieved 86%+ HOTA tracking performance, outperforming 40+ teams in industry competition. | 🍕 Microservices Architecture - Architected distributed real-time hygiene monitoring system using Python, FastAPI, RabbitMQ, and React, achieving 93%+ accuracy in violation detection. Built scalable YOLO-based pipeline with multi-object tracking, state-driven ROI logic, and WebSocket streaming, enabling fault-tolerant processing with sub-second latency updates and comprehensive analytics dashboard. |
| Real-Time Behavioral Analysis System | Multilingual Vision Meets Words |
|
|
|
| 🎯 AI Interview Cheating Module - Developed real-time cheating detection system for AI interviewer detecting 3D head orientation (pitch, yaw, roll) and gaze direction with 99% accuracy. Built low-latency pipeline using MediaPipe FaceMesh, OpenCV, and PnP pose estimation with 3D-to-2D mapping, monitoring multiple subjects simultaneously with zero failures. | 🎓 Graduation Project - State-of-the-art (SOTA) multilingual image captioning system (English, Arabic, Egyptian Arabic) using BLIP, GIT, and ViT-GPT2 with LoRA. Achieved 75.0%/30.0% BLEU scores on Flickr30k and created the first Egyptian Arabic captioning dataset with 8,000+ translated captions, outperforming benchmarks by 15%. |
| Arabic Sign Language (ArSL) Detector | Smart Garbage Classification System |
|
|
|
| 🌟 Accessibility Innovation - Architected a YOLOv10-based Arabic sign language detector with advanced augmentations, achieving 98% mAP. Built real-time OpenCV pipeline converting hand signs into live Arabic text via custom overlay, making communication more accessible for the Arabic deaf community. | ♻️ Environmental Impact - Established and deployed high-accuracy waste classification model (95%) categorizing materials like plastic, glass, cardboard, paper, trash, and metal. Built comprehensive solution promoting sustainable waste management and environmental responsibility. |