I'm an engineering student at Cairo University focused on building practical LLM-powered systems. My work sits at the intersection of AI agents, RAG pipelines, and real deployment — not just notebooks.

Right now I'm building:

• Agentic systems with LangChain that use tools, maintain memory, and handle multi-step reasoning
• RAG pipelines for making private or technical documents queryable in natural language
• End-to-end AI applications that go from problem → working implementation

I care about building things that actually run, not just things that look good in a demo.

Core

Python · LangChain · OpenAI API · LangGraph (learning)

AI & LLMs

RAG · LLM Agents · Tool Use · Memory Management
Prompt Engineering · ChromaDB · HuggingFace · Embeddings
Mistral · GPT-4o-mini · Ollama (local inference)

ML & Data

PyTorch · XGBoost · Scikit-learn
Pandas · NumPy · Matplotlib · Seaborn

Tools & Deployment

Git · Gradio · FastAPI (learning) · Docker (learning)
Bash/Shell · UNIX/Linux · VS Code · Jupyter

LangChain · OpenAI API · Python · DuckDuckGo Search

Terminal-based LLM agent that analyzes CV-to-job fit end-to-end. Give it your CV and a job description — it searches the web for company context, scores your CV against each requirement, identifies gaps with priority order, and builds a prep plan. Supports multi-turn follow-up questions with persistent memory.

What it demonstrates: custom tool design · LangChain agents · conversation memory · prompt engineering

LangChain · ChromaDB · Mistral 7B · HuggingFace · Gradio · Bash

End-to-end RAG system for querying technical EDA documentation (Calibre) in Arabic and English. Uses multilingual-e5 embeddings, ChromaDB with MMR retrieval, and local Mistral 7B inference via Ollama — zero API cost. Includes a Gradio interface, query logging, and a Bash deployment script for Linux.

What it demonstrates: RAG architecture · vector search · local LLM deployment · evaluation

Python · Scikit-learn · NASA Datasets · REST API

Built at NASA Space Apps Hackathon (48hrs). End-to-end ML pipeline from satellite data ingestion to deployed API powering a public forecast website. Handled multi-parameter environmental data: PM2.5, PM10, NO2, O3.

What it demonstrates: full pipeline under pressure · API deployment · real data challenges

• Walmart Sales Forecasting — XGBoost + time-series feature engineering
• Loan Approval Classifier — 97% accuracy on imbalanced dataset
• Customer Segmentation — K-Means + DBSCAN with Silhouette evaluation

• Machine Learning Specialization — Stanford / Coursera
• Deep Learning Specialization — DeepLearning.AI
• Mathematics for ML and Data Science — DeepLearning.AI
• HuggingFace LLM Course (in progress)