AI Engineer building production systems across LLMs & RAG, Computer Vision, and Full-Stack · from prompt design through deployment and live incident response.
📍 Cairo, Egypt · 🎓 BSc Artificial Intelligence · 📄 2× IEEE-published · 🏅 National AI Olympiad · 2nd place
I'm an AI Engineer and full-stack builder with two IEEE-published deep-learning papers and a national AI Olympiad medal (2nd place) across Computer Vision, Embedded Systems, and Cybersecurity.
As the first and sole engineer at a Saudi ride-hailing operator, I designed, built, and operate its entire internal software stack · roughly 15 production systems spanning customer support, CRM, AI driver-screening, incentives, dispatch, and internal communications · while coordinating and mentoring a 3-person operations team.
I work deep across LLMs & RAG, computer vision, and production full-stack engineering (TypeScript, Python, React, Next.js, PostgreSQL), with a test-heavy, ship-to-production mindset: migrations, deployments, API rotations, and live as-a-user verification before every release.
Real, high-scale systems running live in production. Enterprise & internal platforms are proprietary · shown here with architecture and hard metrics rather than public links.
🚗 As the first and sole engineer at SHEGO (a Saudi women-only ride-hailing operator on the ONDE platform), I designed, built, and operate the entire internal software stack · roughly 15 Arabic-first production systems · owning architecture, 160+ database migrations, deployment, API-key rotation, and live incident response, while mentoring a 3-person ops & content team. A selection below.
End-to-end help desk and CRM ingesting WhatsApp & email into tickets, with an advisory-lock auto-assignment & routing engine, SLA escalation, a 84-toggle role-based access model, and a 27-section quality-evaluation system.
Vision-language document extraction from Saudi licenses, national IDs & iqamas (Gemini + GPT-4o), Arabic-name gender fusion, and a WASL/TGA compliance decision engine verified row-for-row against the official Saudi spec. Caught real identity fraud.
A GPT-4o vision pipeline that reads IBAN, amount & beneficiary from documents and matches them deterministically, plus a confidence-gated AI auto-reply system with per-customer memory and a token usage-cost ledger.
A money-critical incentive & retention system with an append-only ledger and serializable transactions enforcing trip caps under concurrency, plus a Gemini-driven smart-surge engine issuing real driver payouts.
Demand forecasting with Poisson modeling over Uber H3 geospatial zones, a forecast-accuracy A/B ledger, and a Google Maps operations console · plus an automated leads pipeline ingesting Meta, TikTok & Snapchat lead ads into WhatsApp recruitment templates.
A custom Slack/Teams with real-time LiveKit audio & video for up to 100 participants, plus a CRM replacing Zoho (27 features on n8n), a request-lifecycle tracker (Postgres triggers + Gemini subtask splitter), a tasks & approval-chains system, a referral workspace, and a training-quiz platform.
Stood up AABAR Mentor as a separate Go + React deployment on a DigitalOcean droplet with its own database, preserved SSO and a GPT-4 flow · live over HTTPS as a client-facing AI coaching product.
Contributed to 7 production AI services · audio transcription, email auto-reply drafting, social-media automation, custom AI agents, AI website builders, Shopify product automation, and smart visual insights · and shipped n8n workflows for daily content publishing (Google/LinkedIn/OpenAI) + Copilot Studio agents.
An Attention U-Net + EfficientNet-B3 segmentation model reaching 0.75 mean Dice across 1,588 images / 169 patients, with patient-level splits and mixed-precision training, plus a clinical inference pipeline with Grad-CAM explainability and structured measurement extraction.
Public projects with full source on GitHub · computer vision, RAG agents, NLP and ML.
Detects aircraft body type, jet-bridge and boarding-stairs connections from airport gate video. YOLOv8 segmentation + Groq LLaMA-4 Scout zero-shot vision + Streamlit UI.
Chat-first GCC product-compliance assistant: deterministic 3-stage decision engine, bilingual EN/AR chat, OCR pipeline · $0 per query. FastAPI + vanilla JS.
LangChain orchestrator with intent classification, a RAG sub-agent (FAISS + text-embedding-3-small) for menu/policy queries, and an operations sub-agent with MCP-style tool calling for bookings & specials.
Pipeline producing training-ready synthetic Egyptian-Arabic speech: LLM prompts → multi-voice TTS → automated Whisper QA → human review → HuggingFace export.
Real-time heavy-equipment activity tracking prototype · YOLOv8 detection, custom hybrid tracker, masked optical flow, Kafka streaming, PostgreSQL & a Streamlit dashboard.
Deep CNN models for plant-disease classification (the topic of my IEEE CSDGAIS 2024 paper on a novel CNN with L1 regularization). Includes an enhanced ensemble and a ResNet variant.
Transformer fine-tuning for sentiment & emotion detection using transfer learning and embedding extraction.
Kalman filtering and signal integration across multiple sensor inputs for robust vehicle-distance estimation.
Collaborative-filtering recommendation engine built on customer reviews and purchase behavior.
CNN models for product-image recognition and a visual-similarity recommender, built from product-image datasets (Slash Hub).
End-to-end ML pipeline predicting freight shipping rates from historical logistics data · feature engineering, model training and evaluation.
Supervised ML on imbalanced transaction data to flag fraudulent charges · EDA, resampling, and multiple classifier comparison.
Classic diagnostic ML models predicting heart-disease and breast-cancer risk from clinical features, with model evaluation and interpretation.
Publicly accessible AI apps and automation workflows you can try right now.
A real transformer for sentiment & emotion (English & Arabic) running 100% in-browser via Transformers.js · no server, no API key. A live, public Hugging Face Space.
Open live demo →Upload any photo and a DETR transformer draws labeled bounding boxes · real computer vision running entirely in your browser. Ties to my YOLOv8 detection work.
Open live demo →Upload a photo and a vision-language model writes a natural-language caption · in-browser, no server. Mirrors my production VLM / document-AI work.
Open live demo →A published collection of production-style n8n automation workflows · AI content pipelines, lead-capture automations and LLM agents · with importable JSON, diagrams and docs.
Browse workflows →Sustainable agriculture & food-system resilience · 98.47% accuracy with Grad-CAM explainability. Published on IEEE Xplore.
Deep CNN architecture with L1 regularization for robust plant-disease classification.
Across Computer Vision, Embedded Systems, and Cybersecurity.
Transformer-Based NLP (NVIDIA) · AWS ML & Cloud Foundations · Advanced Learning Algorithms (DeepLearning.AI & Stanford) · Data Scientist Pro & Deep Learning with PyTorch (DataCamp) · Meta Front-End.
Open to AI Engineer / Full-Stack roles · remote or on-site. Available immediately.