Description

I help companies design and deploy production-ready AI systems based on Large Language Models (LLMs), enabling them to turn complex data and internal knowledge into actionable, automated and reliable workflows.

I specialize in building end-to-end LLM solutions such as RAG systems, multi-agent architectures, and AI assistants integrated into real business environments. My focus is on delivering systems that are not just prototypes, but scalable, observable, and deployable in production.

With experience in industrial environments and enterprise AI platforms, I can help teams bridge the gap between AI models and real operational use cases: knowledge retrieval, engineering automation, decision support systems, and internal AI tools.

I typically work on projects involving LLM-based applications, agentic workflows, enterprise RAG systems, AI automation platforms, and production deployment of machine learning models with monitoring and reliability constraints.

Langues

Français
Bilingue ou natif
Anglais
Bilingue ou natif

Préférences en matière de lieu de travail

Accepte de travailler sur site

Paris (jusqu’à 30 km)

SLB
AI Engineer
ENERGIE
août 2024 - mars 2026 (1 an et 7 mois)
Boston, MA, USA
Enterprise Multi-Agent AI Platform for Industrial Engineering.

Designed and deployed a production-grade multi-agent AI platform to assist engineers, researchers, and technical experts in the oil & gas industry.

The challenge was to provide unified access to large volumes of confidential technical knowledge while enabling users to perform advanced engineering analyses directly from a conversational interface.

Key capabilities included:
Knowledge agents connected to enterprise document repositories through Retrieval-Augmented Generation (RAG), providing source-grounded answers with full traceability.
Engineering agents capable of executing statistical models, machine learning algorithms, and domain-specific analysis workflows directly from natural language requests.
Tool-enabled agents connected to internal engineering systems and simulation software through MCP and custom integrations, allowing the platform to trigger analyses and operational workflows.
Human-in-the-loop validation processes for critical engineering decisions, ensuring reliability and compliance with operational requirements.
End-to-end reasoning traceability, providing visibility into agent decisions, executed tools, retrieved sources, and workflow execution paths.
To meet strict confidentiality requirements, the platform was deployed entirely on-premise using open-source LLMs (Llama, Mistral), vector databases, Kubernetes, and enterprise infrastructure.

Technologies:
LangGraph, LangChain, LangSmith, Multi-Agent Systems, RAG, MCP, Kubernetes, Docker, Ollama, Llama, Mistral, Vector Databases, Python, flask, React

Business Impact:
Significantly reduced time spent searching across technical documentation.
Enabled engineers to access domain expertise and engineering tools through natural language interactions.
Unified knowledge retrieval, AI-powered analysis, and engineering workflows within a single platform.
Python LangGraph LLM Agents IA RAG
Farm3
AI Engineer
AGROALIMENTAIRE
février 2024 - juillet 2024 (5 mois)
Paris, France
AI-Powered Agronomic Prediction System

Developed production-ready machine learning systems leveraging hyperspectral imaging to support agronomists and farmers in crop health monitoring and decision-making.

The objective was to transform large volumes of hyperspectral data collected from drones and field sensors into actionable insights capable of detecting plant stress, nutrient deficiencies, and crop health issues at an early stage.

My role covered the entire machine learning lifecycle, from data ingestion and model development to deployment and operational monitoring.

Key contributions:

Designed and implemented end-to-end ML pipelines handling hyperspectral data ingestion, preprocessing, training, validation, and deployment.
Developed predictive models based on high-dimensional spectral data, combining feature engineering, clustering techniques, and machine learning algorithms to improve predictive performance.
Optimized XGBoost models on hundreds of spectral bands, achieving a 15% improvement in prediction accuracy.
Built monitoring workflows and operational dashboards to track model performance, prediction quality, and business KPIs.
Developed user-facing applications enabling non-technical agronomists to interact with predictions and explore crop health indicators in real time.
Implemented reproducible deployment workflows using Docker and MLOps best practices.
Technologies:
Python, Scikit-Learn, Pandas, NumPy, Docker, Streamlit, MLOps

Business Impact:
Improved prediction accuracy by 15% on hyperspectral crop analysis tasks.
Enabled earlier detection of plant stress and agronomic risks.
Reduced the gap between AI models and field operations through intuitive visualization tools.
Delivered production-ready ML systems directly usable by agronomists and operational teams.
Machine learning Python MLOps Data Engineering Modélisation prédictive
Aubay
Data Scientist
CONSEIL & AUDIT
février 2023 - novembre 2023 (9 mois)
Paris, France
Multimodal Deep Learning System for Sign Language Translation

Developed an end-to-end AI system aimed at improving accessibility for deaf and hard-of-hearing users by automatically translating French Sign Language (LSF) videos into text.

The project addressed the lack of existing high-quality datasets and production-ready solutions in this domain, requiring both data engineering and model design from scratch.

Key contributions:

Led the full lifecycle of the project, from problem definition with stakeholders to production deployment of the final application.
Built a custom data pipeline for video-based sign language data, including data collection coordination, annotation strategy, and preprocessing of temporal-spatial features.
Designed a multimodal deep learning architecture combining Convolutional Neural Networks (CNNs) for spatial feature extraction and Transformers for temporal sequence modeling.
Trained and optimized the model to capture complex dependencies between hand shapes, motion dynamics, and sequential sign structure.
Achieved approximately 85% accuracy on test sequences, validating the model’s ability to generalize across diverse sign patterns.
Developed and deployed a web-based application to make the model accessible to end users and external stakeholders.

Technologies:
Python, PyTorch, CNN, Transformers, Computer Vision, Deep Learning, Docker, flask, html, css, JavaScript

Business Impact:
Enabled automatic translation of sign language videos into text for accessibility use cases.
Improved accessibility of digital content for deaf and hard-of-hearing communities.
Deep Learning Pytorch Python Docker NLP