Work

• Architected and built a multi-tenant marketing-automation platform end-to-end across frontend, backend, and infrastructure
• Built the analytics pipeline powering live engagement dashboards and dynamic segments, sustaining thousands of events/second at peak.

• Shipped a self-service GCP VM platform with custom RBAC and a full audit log that cut cloud spend 40%.
• Migrated an OCR service from Flask to FastAPI with request batching, improving processing speed.

• Owned a platform automating music distribution to Spotify, Apple Music, Anghami & YouTube, replacing a manual, bus-factor-heavy process.
• Built a Scrapy + Selenium crawler for piracy detection, pulling 100+ infringing URLs a day.

Mandatory national service.

• Built cross-framework compatibility layers in Ivy so PyTorch code runs unmodified on a JAX backend, covering core linear-algebra and vision ops.
• Consolidated brittle internal automations into one CI/CD-deployed repo on GCP Cloud Run.

• Built a PyQt desktop app that extracts fields from Egyptian national ID cards.

• Generated a 200K-image synthetic dataset of Arabic and English scene text with TRDG and SynthText, patching both tools' source to fix Arabic glyph rendering and expand font and background diversity.
• Benchmarked state-of-the-art text-detection models on synthetic and real-world data, writing converters to reformat labels for each architecture; the work fed an Arabic scene-text detection research paper.

A distributed key-value store with the Raft consensus algorithm implemented from scratch.

My bachelor thesis: automating analog transistor sizing with genetic algorithms.

A pipeline ingesting the 25M-rating dataset into BigQuery, from data lake to dashboards.

Added a wrapper enforcing correct environment initialization and reset order, merged into main and enabled by default across every Gym environment.

Contributed an example showing distributed, multi-GPU fine-tuning of Hugging Face transformers on a custom text-classification dataset.

Fine-tuned a vision transformer for early detection of rice blast disease from crop images.

Trained an XGBoost model to reconstruct missing GPS readings in a running-tracker app.