Loay Chlih
AI Researcher & Engineer
👋 Hi, I’m Loay
I’m an AI Engineer who recently completed a final-year internship at Veepee in Paris, France, where I worked on benchmarking and evaluating Large Language Models (LLMs) for real-world business use cases. I hold a Master’s in Natural Language Processing from the University of Nantes and a degree in Computer Science Engineering from ENSIAS. I am passionate about Generative AI, NLP, and building scalable AI tools that solve practical problems.
I’m actively seeking a position in Generative AI, where I can contribute to research and development in LLMs, evaluation methods, and responsible AI systems.
Education
-
Master’s in NLP University of Nantes (Sep 2024 – Jun 2025) -
Engineering Degree in Computer Science ENSIAS, Rabat (Sep 2021 – Jun 2024)
Work Experience
AI Engineer Intern @ Veepee (Ventes Privées), Paris (Feb 2025 – Aug 2025)
Designed and developed “Model Arena”: a platform for evaluating and comparing LLMs on use cases specific to Veepee.
- At Veepee, any decision about LLMs now goes through this platform. It’s been adopted by several tech teams and has become the go-to standard for making reliable, traceable, and comparable LLM choices.
- Designed of a scalable backend with message queues and asynchronous workers to orchestrate multi-provider API calls (OpenAI, Gemini, etc.) using a system of dynamic templates and structured output formats.
- Built an LLM-as-a-Judge evaluation pipeline to automate qualitative output assessments.
- Integrated flexible evaluation metrics (accuracy, cost, token usage).
- Tools: Python, FastAPI, PostgreSQL, Docker, React, LLM APIs.
AI Research & Development Assistant @ AI Movement, Rabat (Feb 2024 – Jul 2024)
Integrated ML into Branch-and-Bound algorithms for solving TSP.
- Creation of a pipeline for generating TSP instances and benchmarking machine learning models.
- Improved TSP solution time by integrating a GCNN model: gains of 11.44% for TSP-15, 16.24% for TSP-20, and 10.26% for TSP-25.
- Tools: PyTorch, SCIP, Numpy, Docplex, Weight&Biases.
Data Science Intern @ SQLI, Rabat (Jul 2023 – Sep 2023)
Built a web-based sentiment analysis pipeline for e-commerce reviews.
- Data extraction through web scraping and benchmarking of sentiment analysis models to identify the best-performing one.
- Integration of the best model as an API, enabling users to analyze their data via a web page.
- Tools: PyTorch, Hugging Face Transformers, NLTK, Streamlit.
Projects
Synthetic Data Generation for an E-commerce Chatbot use case
GitHub: Agent_E-Commerce-Chatbot
A powerful synthetic data generation pipeline using the Ragas framework to simulate realistic e-commerce queries and personas.
Key Features:
- LLM-based query classification and knowledge graph construction
- Synthetic data pipeline with 6 customer personas
- Evaluation-ready test sets using Ragas
Tech: Python, Flask, Ragas, Hugging Face, LangChain, FAISS
Structured RAG Hotel Search Chatbot
GitHub: Structured-RAG-Hotel-Search
An AI-powered hotel search chatbot that uses structured query parsing + semantic search to return smart hotel recommendations. Combines FastAPI and React with OpenAI and Qdrant for a real-time Chatbot.
Key Features:
- Extracts structured output (location, services, amenities) using GPT-4o
- Ranks hotels with semantic + filter-based matching
- Frontend shows personalized hotel cards with filters
Tech: FastAPI, React (TypeScript), Qdrant, OpenAI GPT-4o
Languages
- French (fluent)
- English (fluent)
- Arabic (native)