mohamed lotfy

At AHBS, I build R&D AI systems including LLM agents, code graphs, database agents, and a speech diarization pipeline for long-form audio and video.

I engineered an AST-backed code graph with FastAPI and Neo4j that reduced documentation and code comprehension effort by 82%, and deployed LLM-powered search, summarization, and document-analysis solutions integrated with Microsoft Teams and SharePoint. I also developed a quality-gate classifier achieving 91% accuracy and a pharmacist assistant for active-ingredient analysis and alternatives.

Previously at Orange Innovation, I optimized a deployed AI summarization service, improving efficiency and accuracy by 10% while reducing inference time and operational costs by 5%. I also built a real-time screen-aware LLM assistant and earned a Bronze Medal in the Kaggle LLM Science Exam using LoRA, QLoRA, and Mixture of Experts.

I teach AI, machine learning, data engineering, computer vision, reinforcement learning, and LLM architectures through the National Technology Institute and Digital Egypt Pioneers Initiative. My research includes Arabic speech-to-sign language, Arabic text recognition, and LLM evaluation, supported by hands-on work with PyTorch, TensorFlow, RAG, FastAPI, Flask, and MLOps.