I'm an AI Engineer who builds real-time computer-vision systems that ship — from sign-language translators running on web and mobile, to multi-camera tracking platforms streaming through Kafka, to live pose-estimation coaches that score human movement on the fly. My focus: turning research-grade CV and generative-AI models into production software with measurable impact.

• 🔭 Currently building a real-time construction-equipment tracking platform (YOLOv11 + BoT-SORT + Kafka + TimescaleDB).
• 🧠 Deep interest in multimodal AI, vision-language models, pose estimation, and edge inference.
• 🎓 B.Sc. Computer & Control Systems Engineering — graduation project graded Excellent.
• 💬 Ask me about: YOLO pipelines, MediaPipe pose estimation, optical-flow analytics, FastAPI services, real-time streaming, or deploying CV models on edge devices.

Languages

AI / Computer Vision

Backend & Data

Apps & Tooling

End-to-end CV platform that detects, tracks, and analyzes heavy machinery on-site in real time.

YOLOv11 · BoT-SORT · Optical Flow · Kafka · FastAPI · TimescaleDB · Streamlit

Streams multi-camera feeds through a Kafka pipeline, runs YOLOv11 detection + BoT-SORT identity tracking, derives idle/active states from optical-flow motion analysis, persists time-series telemetry to TimescaleDB, and surfaces a live operations dashboard. Built for real construction-site reliability constraints, not lab demos.

Real-time ArSL ↔ Arabic translator with 3D-avatar synthesis. Graduation project — graded Excellent.

YOLOv8l · Auto-Correction NLP · 3D Avatars · Web · Mobile

Recognizes Arabic Sign Language gestures via fine-tuned YOLOv8l, post-processes outputs through a context-aware auto-correction module to produce coherent Arabic sentences, and renders the reverse direction (text → sign) using animated 3D avatars on web and mobile clients.

Real-time fitness coach that tracks form, counts reps, and scores performance from a single webcam.

MediaPipe Pose · OpenCV · Streamlit · Python · NumPy

Builds a real-time AI fitness coaching system by streaming live webcam frames through MediaPipe Pose for 33-landmark body tracking, driving per-exercise rep state machines on joint-angle thresholds, and classifying form errors in real time across squats, push-ups, and bicep curls. Each session emits a 0–100 performance score, JSON telemetry, and a Streamlit analytics dashboard with cross-session comparison.

Production-ready desktop ERP for pharmacies.

Python · PyQt6 · SQLite · Vector DB

Inventory, employees, products, and image-backed catalog management with a modern PyQt6 UI. Architected for an AI upgrade path — vector-database-powered semantic search and recommendations.

I'm open to AI Engineer / Computer Vision / ML Engineer roles and collaborations on real-time vision and generative-AI systems.

• 📧 Email: karimelroby20@gmail.com
• 💼 LinkedIn: Karim Elroby
• 🐙 GitHub: @KarimElroby7
• 📊 Kaggle: @karimelroby7
• 📍 Cairo, Egypt — open to remote & relocation