For Group Activity Recognition V2, I designed VAT-Former to recognize volleyball team activities from player body keypoints. The system achieved 91.10% Top-1 accuracy and 91.07% F1 score with 883K parameters.
I built the computer vision backend for AI-Powered Fitness Coach, which counts reps, detects form errors, and gives voice feedback in real time. Its evaluation system combines geometric state machines, a VAE, and an STGCVAE.
For SmartClassAI, I built a CUDA-accelerated computer vision pipeline for student identification, attendance, and security alerts. I also developed a guided biometric enrollment flow and a web layer with role-based portals.
At DigitalEgyptPioneersInitiative(DEPI), I built Pathora, an AI career assistant that parses CVs, predicts job fit, generates adaptive exams, evaluates answers with an LLM, and recommends courses. I also teach Python, mBlock, and VEX Robot coding at iSchool.

