I am an AI Engineering Major at Mansoura University (GPA 3.82/4.0) specializing in deploying and optimizing LLMs for scalable, high-performance retrieval and generation tasks. My expertise spans fine-tuning state-of-the-art models and building robust Retrieval-Augmented Generation (RAG) systems.
- Fine-Tuning: Experienced in low-rank adaptation methods (QLoRA, LoRA) for resource-efficient model customization.
- RAG Architecture: Focused on advanced RAG patterns, including hybrid search (semantic + keyword) and handling massive, long-form documents.
Samsung Internship (LLM Engineer)
- Fine-tuned Llama 3.1 (8B) using QLoRA and 4-bit quantization on proprietary pricing data.
- Result: Achieved 75% accuracy in complex product price prediction, significantly improving internal forecasting models.
Freelance Data Scientist (Automation Pipeline)
- Developed automated book summarization pipelines leveraging LangChain and chunking strategies to handle long-form content that exceeds standard context windows.
| Domain | Technologies |
|---|---|
| LLM & RAG |     |
| Deep Learning |    |
| Core & DevOps |     |
A high-performance RAG system built to manage and retrieve knowledge from 4,000+ HR documents.
- Architecture: Built with FastAPI, LangChain, and Qdrant for vector indexing, backed by MongoDB.
- Impact: Engineered a solution that achieved 87% retrieval accuracy while maintaining sub-2-second query latency under production load.
- Innovation: Implemented a unique dual-database architecture using Sentence-Transformers (384-dim) for optimized semantic search and retrieval speed.
An intelligent, web-scraping tool that uses advanced LLMs to generate structured summaries from lengthy website content.
- Tech Stack: Gradio UI, BeautifulSoup for scraping, and multiple LLMs via OpenRouter (e.g., Qwen, Deepseek).
- Feature: Designed a modular architecture with streaming responses, allowing easy switching between various OpenRouter-hosted models without code changes.