Freelancer profile translated to English.

Description

I design and industrialize generative AI agents.

LLM / RAG agents that query your document corpora with sourced answers, draft business deliverables, and integrate with your platforms — end-to-end support: architecture, development, deployment, evaluation. Including with self-hosted open-weight LLMs when your context requires it.

PROJECT TYPES

RAG agents and documentary assistants — sourced and traceable answers on your corpora
Business assistants: documentary analysis, assisted drafting of deliverables
Industrialization of your AI POCs: evaluation, reliability, monitoring, production deployment
Scoping in workshops, training, and upskilling your teams

AREAS OF EXPERTISE

LLM / RAG Agents — LangGraph, LangChain, vector search (Milvus), reranking, prompt engineering, evaluation (RAGAS, LangSmith, Langfuse)
Production & Industrialization — Dataiku, MLOps / LLMOps, monitoring, robustness
Machine learning & NLP — predictive models, semantic analysis, classification

REFERENCES

Caisse des Dépôts (generative AI agents in production), Bouygues Telecom (national scale ML, +10M customers), La Poste, Business & Décision — Orange group. Over 6 years of data science and machine learning in production.

Dataiku Expert: 6 certifications, including Generative AI Practitioner and MLOps Practitioner.

Have a generative AI or machine learning project to put into production? Let's talk.

Industry field of expertise

Languages

French
Native or bilingual
English
Fluent

Workplace preferences

Can work on-site

Paris (up to 50km)

Groupe Caisse des Dépôts
Data Scientist — Generative AI
BANKING AND INSURANCE
November 2025 - Today (8 months)
Paris, France
Design, development, and production deployment of generative AI agents applied to governance, in a sovereign environment: self-hosted open-weight LLM, strictly internal data, no external access.

GOVERNANCE INSTANCE PREPARATION ASSISTANCE AGENT
Multi-tool conversational agent querying the instance document corpus (board of directors, audit and risk committees…), with sourced and traceable restitution of each answer.
Iterative drafting of analysis notes: successive versions kept throughout the exchange, with an update logic controlled on the code side and drafting delegated to the model.
Strict compartmentalization of sensitive data — independent controls applied to each documentary call, ensuring isolation between perimeters.
Production deployment on the Dataiku platform, with response streaming and stateless architecture focused on robustness.
GOVERNANCE PROJECT SI PROCEDURES NAVIGATION ASSISTANT (RAG — POC)
Natural language querying of voluminous procedures (>100 pages), with sourced answers tailored to the user's profile.
Retrieval + reranking pipeline, iterative prompt engineering.
Needs scoping in workshops with business teams; deployment on Dataiku Agent Hub.
TECHNICAL STACK:
- Agents & orchestration: LangGraph, LangChain, tool / function calling, multi-tool orchestration
- Retrieval: vector search, Milvus, multilingual embeddings, reranking, semantic routing
- LLM: self-hosted open-weight, advanced prompt engineering (few-shot, counter-examples), guardrails, evaluation & reliability of answers
- Platform & engineering: Dataiku (Agent Hub, Knowledge Bank, Code Agent), Python, stateless architecture, streaming, batch processing, Git / GitHub
Generative AI LLM RAG AI Agent Dataiku
Bouygues Telecom
Data Scientist / ML Engineer
TELECOMMUNICATIONS
October 2022 - October 2025 (3 years)
Meudon, France
Transforming technical network signals into customer experience predictions: 3 years designing and industrializing the Customer Experience team's ML models (Radio Department), operated at national scale.

CUSTOMER SATISFACTION & CHURN
Predictive model for national customer satisfaction and network quality (over 10 million customers), built on core network probe data with a Multi-Touch Attribution approach.
Daily and monthly scoring identifying the top 1% of most dissatisfied customers, directly feeding team retention actions.
NETWORK PERFORMANCE
Model for anticipating network performance drops and detecting traffic peaks, to act upstream of customer experience degradation.
TECHNICAL STACK:
- ML & data: Python, PySpark, scikit-learn, XGBoost, Isolation Forest, NumPy, Pandas
- Industrialization: Dataiku, Airflow, GitLab CI/CD, Git, Amazon S3
- Data & visualization: SQL, Tableau
Machine learning Python MLOps Scikit-learn Dataiku
BUSINESS & DECISION
Data Scientist Consultant
DIGITAL AND IT
September 2020 - August 2022 (1 year and 11 months)
Nantes, France
Data science consulting within the Orange group: upskilling teams and building end-to-end AI products — from web scraping to production deployment.

TRAINING & SUPERVISION OF A DATA SCIENCE TEAM
Technical training and supervision of a team of 5 data analysts (Python, Dataiku): upskilling in NLP, from preprocessing to model training (sentiment analysis, topic modeling / LDA).
AI-POWERED HR TOOL FOR RECRUITMENT (R&D)
Design of an ATS-type tool to aid recruitment and mission assignment: approximately 2000 internal CVs processed by the tool, model trained on several thousand web-scraped profiles.
Complete AI cycle: web scraping, semantic annotation, Named Entity Recognition, model deployment and monitoring.
BI & REPORTING
Power BI dashboards and reporting; migration of reports to SAP BO (Informatica ETL), automation of reporting generation.
TECHNICAL STACK
- NLP & ML: Python, scikit-learn, spaCy, NLTK, Spark-NLP, Gensim, PyLDAvis, DeepPavlov, NER, sentiment analysis, topic modeling (LDA)
- Platform & tools: Dataiku, Git, GitLab, Power BI, Tableau, SAP BO, Informatica
NLP Sentiment Analysis Web Scraping Training Microsoft Power BI