AI & Data Analytics Engineer — computer vision, retrieval-augmented generation and LLM systems.
Cairo, Egypt · Open to AI / Data Analytics roles
I build systems that turn messy data and raw video into decisions people can act on. My work sits where machine learning meets the unglamorous parts: evaluation harnesses, reliability layers above a model, deployment that fits in 512 MB, and tests that catch the bug before a user does.
I care about the part after the demo. A letter-by-letter sign classifier is easy; making one that does not fire on every hand movement is the actual engineering.
| Project | What it is |
|---|---|
| SignBridge | Real-time sign-language translation in the browser — sign ↔ text ↔ audio. Arabic (33 classes) and American (28 classes) sign, with WebRTC rooms, a 3D avatar and an interactive learning mode. Graduation project. |
| PPE Detection (YOLOv11) | Workplace-safety compliance detection. YOLOv11 fine-tuned for reflective vests and hard hats, served through a Streamlit dashboard for images, video and live webcam. |
| Nexus RAG | Bilingual retrieval-augmented generation. BM25 and dense vectors fused with Reciprocal Rank Fusion, answering in Arabic or English with page-level citations. |
| SignBridge AI | Sign-language ML pipeline over WLASL — dataset subsetting, video I/O and a pluggable pose backend — with a React/TypeScript interface. |
| Qwen3-4B Business Copilot | A 4B-parameter model fine-tuned for a business domain and published on the Hugging Face Hub under Apache 2.0. |
The parts of the work that rarely show up in a screenshot:
- CI on every repository — lint, tests, dependency audits, and, for the computer-vision repo, a job that fails the build if the dataset is ever committed again.
- Tests that run in seconds — suites written to need no API keys, no database and no GPU. 17 tests for PPE, 31 for RAG, 39 for the ML pipeline.
- Bugs found by writing the tests. The RAG loader had a fallback for non-UTF-8 files that never fired, because LangChain re-raises
UnicodeDecodeErroras aRuntimeError. Arabic summary detection matched inside longer words. Both were invisible until something asserted them. - Product-minded analytics — built 10+ Power BI dashboards over 15 KPIs and cleaned and analysed 30+ datasets during a 300-hour applied programme.
| Certification | Issuer |
|---|---|
| Power BI Data Analyst Associate (PL-300) | Microsoft |
| AI Engineering Professional Certificate | IBM |
| Data Analytics Professional Certificate |
Languages Python · SQL · TypeScript · JavaScript
AI/ML PyTorch · scikit-learn · TensorFlow / TFLite · Ultralytics YOLO · Hugging Face Transformers · MediaPipe · LangChain · Unsloth
Data & backend pandas · NumPy · scikit-learn · FastAPI · Streamlit · Qdrant · BM25
Frontend & BI React 19 · Vite · Tailwind · Radix UI · Three.js · WebRTC · Power BI · Tableau
Practice pytest · Ruff · GitHub Actions · Docker · Render
- LLM evaluation frameworks and production guardrails
- Real-time computer vision on edge devices (TFLite / ONNX)
- Bilingual (Arabic/English) retrieval and generation
BSc, Management Information Systems — MCI Academy, Cairo (2019–2023), GPA 3.7/4.0.
Coursework in data analytics, database systems, business intelligence, statistics and information systems management.
Arabic — native · English — professional working proficiency
The fastest way to reach me is LinkedIn or email: [email protected]
I am actively looking for AI / Data Analytics roles where the work has to survive contact with real users. If you are hiring, or you want to talk about retrieval, computer vision or assistive technology, I would be glad to hear from you.