~/portfolio $ whoami
Available for AI engineering rolesAI Engineer · M.Sc. Artificial Intelligence, Cairo University
I ship production AI systems across Computer Vision, NLP, and Generative AI, from data and model development to Dockerized deployment on AWS. My focus is RAG systems, agentic multi-agent workflows, and getting research-grade models into dependable production.
about
I'm an AI Engineer who builds end-to-end systems: data pipelines, model development, evaluation, and production deployment. Recent work spans RAG chatbots, agentic multi-agent workflows (ReAct, LangGraph), OCR-plus-LLM document extraction, and ensemble forecasting models delivered to real clients, including four Jordanian public universities.
My core stack is Python, PyTorch, TensorFlow, LangChain, LangGraph, Hugging Face, FastAPI, and Docker, working with both open and proprietary LLMs (GPT, Mistral, LLaMA, Gemma) through fine-tuning, prompt engineering, and API integration. I'm currently pursuing an M.Sc. in Artificial Intelligence at Cairo University.
Before AI engineering, I spent three years as a competitive programmer, ranking in the top 6 nationally at the Egyptian Collegiate Programming Contest. That's where my instinct for algorithms and problem-solving under pressure comes from. I'm most interested in agentic systems, retrieval-grounded LLMs, and MLOps on AWS (Bedrock, SageMaker).
skills
experience
selected work
Client: Jordanian Ministry of Higher Education (4 public universities)
Four universities needed reliable day-ahead forecasts of on-site solar generation from limited operational history. I built an ensemble regression system (Gradient Boosting, XGBoost, CatBoost, Random Forest) trained on 314–368 days of data, with 30+ engineered time-series features (lags, rolling statistics, cyclical encodings), delivered as an interactive Streamlit dashboard for upload, prediction, and forecast export.
// production & research
A ReAct multi-agent system on Mistral and LangGraph orchestrating 5+ tools (web search via Tavily, email via SendGrid, memory) with dynamic tool selection. Reached ~40% fewer steps to task completion versus a single-agent baseline and sub-3s average latency, shipped as a containerized 3-service Docker deployment.
End-to-end pan-cancer classification on TCGA somatic-mutation data (MAF), building a patient × gene binary mutation matrix with tumor-mutational-burden features across 33 cancer types. Applied SMOTE, XGBoost/Random Forest with SHAP explainability, and per-class Youden threshold tuning, served through a Streamlit front-end.
Production RAG chatbot over a large internal knowledge base using LLaMA 3.2, ChromaDB embeddings, and CrossEncoder re-ranking for precise enterprise Q&A. Automated the ingestion, chunking, and embedding pipeline and deployed it via FastAPI and Docker with CI/CD-ready containerization.
Generative-AI document analyzer using LLaMA 3.2 Vision and EasyOCR to extract structured financial fields (account, amount, date, currency) from statement images into standardized JSON, with a FastAPI backend handling validation and downstream system integration.
Production OCR pipeline that extracts structured data from identity documents, combining YOLOv8 for field detection, EasyOCR for text extraction, and LLaMA Vision for post-processing and correction. Fully Dockerized for scalable, reproducible deployment.
// more projects
XGBoost multi-class classifier estimating obesity level from demographics and lifestyle habits, with real-time interactive predictions and model explanations in Streamlit.
Clinical ML model trained on blood pressure, cholesterol, and ejection-fraction data to estimate heart-failure risk, built to support early, data-driven screening.
K-Means segmentation app where users choose features, find the optimal cluster count via the elbow method, and explore group profiles with auto-generated summaries.
Real-time attendance tracking using FaceNet embeddings and OpenCV with database logging, eliminating proxy attendance through automated face detection and identification.
Sequence-to-sequence LSTM encoder-decoder in TensorFlow that compresses source text into a context vector and decodes the target language word by word from bilingual training data.
recognition
Ranked 6th nationally in three consecutive years (2020, 2021, 2022), demonstrating strong algorithmic problem-solving under pressure.
AI & Robotics leader of the OI ROV Team: 3rd place (2023), 6th place (2024 & 2025), and the No Pain No Gain Award (2022) at the national competitions.
contact
Open to AI engineering roles, collaboration, and project inquiries. If you have an idea or just want to connect, reach out.