MAROUAN BAAZIZ

Project: Design and Development Smart HR Chatbot - Designed the architecture for a Retrieval-Augmented Generation (RAG) pipeline to automate HR query resolution. - Built backend API endpoints using FastAPI and developed a user-friendly chat interface with Gradio for seamless interaction. - Leveraged LangChain, Hugging Face models, and PyTorch to implement the RAG workflow, enabling accurate and context-aware responses. - Used open-source LLMs like Qwen2-72B to generate and evaluate high-quality Q&A pairs with LlamaIndex metrics. - Reduced inference time by 80% through model optimization, vLLM inference, and code refactoring. - Optimized workflow orchestration with LangGraph, enabling complex multi-step reasoning, retrieval loops, and stateful conversation management. - Developed a data extraction pipeline for payroll analysis orchestrated via Dagster, and validated with Pydantic models. - Deployed and monitored the chatbot on OVHCloud (NVIDIA A100 GPU). - Participated in sprint planning, release cycles, and inference quality evaluation to ensure reliable performance and continuous improvement.