Riadh's Personal Site

Selected Personal Projects

The following projects are selected personal initiatives that I develop independently as part of my continuous research exploration and technical experimentation.

These systems are built during my free time to:

  • Explore cutting-edge AI ideas
  • Transform research prototypes into production-ready systems
  • Experiment with modern full-stack architectures (FastAPI + React + Docker + GPU pipelines)
  • Benchmark real-time inference and edge deployments
  • Test new model architectures, biometric pipelines, and interaction paradigms
They are not commercial products, but rather research-driven engineering explorations — where I bridge academic AI concepts with practical, deployable software systems.

FastVLM Web Application

A production-ready full-stack application built on top of Apple’s ml-fastvlm. Includes a FastAPI backend, modern React UI, and Dockerized deployment pipeline — transforming a CLI research model into a usable web product.

FastAPI React Docker Vision-Language Model

PalmID SOTA

A full-stack contactless palm biometric authentication platform with webcam-based enrollment and verification. Built with FastAPI + React, using a hybrid CV pipeline (YOLO11 / MediaPipe fallback + HOG/LBP descriptors) and cosine-similarity matching. Designed for access control and passwordless authentication (note: no liveness detection).

FastAPI React 18 Webcam Capture YOLO11 + MediaPipe HOG/LBP Features Cosine Similarity

Grid Gesture Quiz

An interactive gesture-controlled quiz application that uses real-time hand tracking to answer multiple-choice questions via pinch gestures. Built with Python, MediaPipe, and OpenCV, featuring full Arabic language support and a modern “electric edge blue” glassmorphic UI.

MediaPipe OpenCV Gesture Control Arabic Support Computer Vision Interactive UI

FaceKit Stream

Real-Time Face Detection & Recognition System

A lightweight, database-free real-time face recognition platform supporting webcams, RTSP streams, and IP cameras. Built with FastAPI, OpenCV, and InsightFace, it performs multi-face detection, tracking, enrollment, and cosine-similarity matching — with optional GPU acceleration and full in-memory operation for privacy-focused deployments.

FastAPI OpenCV InsightFace ONNX Runtime Real-Time CV GPU Optional

Motion Detection Project

Production-Ready CPU Video Motion Detection System

A robust, industrial-grade motion detection system optimized for CPU-only environments. Combines background subtraction, optical flow, and camera motion compensation to deliver reliable detection with minimal false positives — even under camera shake, lighting variation, and dynamic backgrounds.

OpenCV Optical Flow MOG2 ORB + RANSAC RTSP Support CPU Optimized

Video Learning App

AI-Powered Video Transcript Translation & Language Learning (English → Arabic)

A web-based language learning system that combines video playback with AI-powered Arabic translations and word-by-word linguistic analysis. Supports local MP4 uploads and YouTube URLs, parses subtitles (VTT/SRT), keeps translations aligned with timestamps, and renders a synchronized dual-panel learning experience.

FastAPI OpenAI (GPT-4o-mini) YouTube (yt_dlp) Subtitle Parsing (VTT/SRT) Word-by-Word Analysis FFmpeg Conversion