Freelancer profile translated to English.

Description

With a strong interest in natural language processing (NLP) since my beginnings in data, and passionate about R&D as well as technological monitoring, I now specialize in language models (LLM) and their concrete applications in business.

I support organizations in the design of artificial intelligence solutions capable of simplifying, or even automating, all or part of their business processes. My approach is based on a strong conviction: the value of AI is fully revealed when it relies on the business expertise of its users. This is why I place co-construction at the heart of each mission.

I work in particular on the development of AI agents (autonomous or orchestrated), the implementation of RAG (Retrieval-Augmented Generation) systems to exploit internal data, as well as on prompt engineering, a real bridge between the business need and the intelligence of the model.

Industry field of expertise

Languages

English
Fluent
Spanish
Basic
French
Native or bilingual

Workplace preferences

Can work on-site

Paris (up to 50km), Reims (up to 50km)

Studeria
AI Trainer
February 2025 - Today (1 year and 6 months)
51100 Reims, France
Creation of training modules in Python and SQL.
Personalized support for executives in their handling of AI and No code tools.
Implementation of workshops aimed at maximizing the use of AI in various professions.
n8n SQL Python No code Pedagogy
BCOME AI Freelance
Data Scientist Consultant
September 2024 - May 2025 (9 months)
51100 Reims, France
Creation of AI agents based on LLMs, capable of performing various tasks (API calls, database querying, etc.) from instructions in natural language Integration of QA pipelines aimed at ensuring the proper functioning of the agents (Django, Python)
LLM Fast API AI Agent Python (Programming Language) RAG
IRT SystemX
AI/NLP R&D Engineer
March 2024 - September 2024 (6 months)
Gif-sur-Yvette, France
Creation of an LLM-based reformulation tool intended to assist engineers in writing specifications for construction equipment for nuclear power plants. The reformulations had to respect a certain number of rules classified by order of importance, while preserving as much as possible the meaning of the initial requirement. Since data confidentiality was of paramount importance, the entire project had to run locally, the LLM as well as the rule compliance verification module were hosted on our machine.

Technical stack: Python, Langchain, FastAPI, Spacy, GitLab, Hugging Face, BERT, Mistral, Ollama, Docker

Key steps:

-drafting of a reformulation protocol with business experts

-creation of the rule compliance verification tool composed of a NER module, several Spacy modules and a SBERT module

-implementation of containerized APIs allowing to call the reformulation module as well as the chosen LLM models

-conducting a state of the art on constrained reformulation with LLM

-prompt optimization with various state-of-the-art techniques at that time (Few-shot, Active Prompting, OPRO)

-implementation of an evaluation protocol by experts, aimed at comparing our tool to the previous tool (transformers T5 trained on a set of 50 examples) as well as reformulations written by engineers
NLP LLM Research and development Mistral Prompt engineering