Mazen Emad

Architected a multi-agent assistant orchestrated with LangGraph on a FastAPI backend, defining agent routing, tool integration, and decision logic to autonomously manage email, calendar, and knowledge-base queries.

• Integrated multiple external tools and APIs (Gmail, Google Calendar, web search, voice STT/TTS) into the agent's toolset, and built a RAG knowledge base over Qdrant for grounded, semantic-search-based generation.
• Implemented a proactive background scheduler that evaluates incoming emails and triggers human-in-the-loop notifications, plus a Pytest test suite and Docker-based MongoDB setup for reliable, repeatable deployment.

• Orchestrated 3 specialized deep learning models (vision, OCR, structured data) through a LangGraph agent state machine, defining task routing and decision logic to autonomously drive a clinical diagnostic workflow end-to-end.
• Implemented one of the pipeline's core agents as a multi-modal fusion architecture in PyTorch — an EfficientNet-B0 CNN backbone (transfer learning) fused with a tabular head for structured lab data — to automate blood-disorder detection, reducing manual analysis time by an estimated 70%.
• Deployed a RESTful API (FastAPI) exposing the multi-agent pipeline for diagnostic inference and automated report generation, achieving 100% uptime under production API constraints — demonstrating the ability to take a multi-agent system from design to a deployed, end-to-end product.

Intelligent WhatsApp integration with Retrieval-Augmented Generation. Processes chat history & documents in real-time. Multi-language support with semantic search and context-aware responses.

Developed a bilingual (Arabic/English) Retrieval-Augmented Generation (RAG) system utilizing LangChain, FAISS vector database, and Ollama (Llama 3 8B) for context-aware document and video analysis.

Engineered an indexing pipeline with dynamic text splitting (512-character chunks) and HuggingFace embeddings (Qwen 0.6B) for semantic search across PDFs and YouTube transcripts. Deployed with Streamlit featuring conversation history management to reduce LLM hallucinations.

Smart farming solution with crop disease detection and yield prediction. Computer vision models identify plant diseases with high accuracy. Data-driven recommendations for optimal agricultural practices.

Architected a real-time facial recognition attendance system using Computer Vision and Deep Learning, processing live camera feeds at 30 FPS with 99% identification accuracy for 500+ personnel.

Engineered secure backend pipelines to manage 10,000+ daily attendance logs and deployed an administrative dashboard for automated report generation, cutting overhead by 80%. Features dual portals for Admin and Students.

An advanced Multi-Agent system powered by CrewAI and Google Gemini. It takes meeting audio, transcribes it using Whisper, and utilizes specialized AI agents to generate summaries, extract action items in structured JSON format, and automatically send follow-up emails.