Thibault Métais

Delivering production-grade AI/ML systems for clients, from architecture to deployment, with a focus on GenAI, multi-agent systems, and scalable infrastructure. Key Achievements: 🚀 Reliable ML Production Pipelines: Built scalable training pipelines and cloud infrastructure integrated with MLflow, enabling reliable, reproducible deployment of client models in production. 🤖 Hybrid ML/LLM & Multi-Agent Systems: Designed hybrid ML/LLM architectures, multi-agent systems, and RAG pipelines, combining traditional models with generative AI to solve client-specific problems. ⚙️ Cost & Performance Optimization: Led infrastructure work on GCP/AWS to balance model performance and cost, including service scaling and containerization improvements. 💻 Tech Stack: Python, PyTorch, MLflow, LLMs, Multi-Agent Systems, RAG, AWS/GCP, Docker, FastAPI.

Transforming experimental city-level models into a global, real-time production system and automating creative workflows with GenAI. Key Achievements: 🚀 Built the "AI Roster" Engine: Led the end-to-end engineering of a core mission-generation system, transforming experimental city-level models into a global production asset. 🌍 Global Predictive Modeling: Built and deployed conversion rate models at a global scale, integrating complex weather & geospatial datasets to maximize prediction accuracy. ⚙️ Industrial MLOps & Architecture: Designed robust pipelines using Airflow, dbt, and SageMaker. Built high-performance FastAPI services to handle real-time mission generation and inference under load. 🤖 GenAI & Process Automation: Developed LLM-based agents to automate artwork workflows. Successfully deployed agents for flyer generation, classification, and analysis, significantly reducing manual operational overhead. 💻 Tech Stack: Python, PyTorch, AWS SageMaker, FastAPI, Airflow, dbt, LLMs.

Building the MLOps Backbone & Standardizing Workflows. Focused on increasing team velocity by building shared tooling and exploring early GenAI capabilities. Key Achievements: 🛠️ MLOps Framework & Standardization: Architected an internal MLOps framework and Python project templates to standardize development. Implemented reproducibility workflows (using DVC & CI/CD) to eliminate "it works on my machine" issues. 🧠 GenAI & R&D Initiatives: Pioneered early RAG prototypes for internal knowledge retrieval. Contributed to advanced analytics workflows including embedding-based transaction clustering and LLM-assisted classification. 🚀 Production & Observability: Deployed robust ML APIs and set up comprehensive monitoring dashboards to track model health in production. 🤝 Leadership & Strategy: Mentored junior Data Scientists on engineering best practices and actively contributed to the technical roadmap planning. 💻 Tech Stack: Python, DVC, CI/CD, LLMs, RAG, Embeddings, Docker.

Core Credit Scoring & Transaction Intelligence. Focused on enhancing risk assessment accuracy through Deep Learning and reliable API deployment. Key Achievements: 💳 Advanced Credit Scoring Built and optimized core credit scoring models, directly contributing to more accurate risk assessment and underwriting decisions. 🧠 Deep Learning for Transactions Developed Neural Networks to analyze complex transaction patterns, unlocking deeper behavioral insights beyond traditional scoring. ⚡ Real-Time Underwriting APIs Designed and maintained high-availability Scoring APIs, serving as the backbone for immediate underwriting decisions. 💻 Tech Stack: Python, Scikit-learn, TensorFlow/Keras, Flask/FastAPI, SQL.

Managed end-to-end projects by defining the mission along with the client, recruiting the student(s) to realize the associated tasks, and track the project progression until its realization.

Developed a web application for fleet managers leveraging localization data to buy the right vehicles for their needs while limiting their environmental impact. Our simulation models enabled them to simulate the use of electric vehicles for their daily operations and plan the corresponding charging infrastructure.