Thomas Pernet

Data Scientist @Groupe Caisse des Dépôts

Paris, FR

I designed and built from scratch a production-grade orchestration system that enables CDC asset managers to query the cross-asset data lake in natural language- The agent can interpret a user’s question, identify the relevant tables and variables, automatically generate and execute SQL queries, run further pandas analysis, and only then deliver the final answer or visualization (JavaScript graph)- The orchestration engine is custom-made using Domain-Driven Design and a Directed-Acyclic Graph (DAG) workflow, without LangChain or LangGraph. It supports dynamic branching (success / empty / failure flows) and produces typed, auditable artifacts (SQL, DataFrames, charts, text)- In parallel, I built the chatbot application from scratch with Flask + HTML/JavaScript, providing asset managers with a user-friendly interface to query and visualize results.This delivers a robust, transparent, and extensible foundation for LLM-assisted analytics in financial data management.

University of Paris I: Panthéon-Sorbonne

Doctor of Philosophy - PhD

北京经济管理职业学院 Beijing Institute of Economics Management

Linguistic program

Ecole normale supérieure

Master’s Degree

University of Paris I: Panthéon-Sorbonne

Master's degree

Fudan University

PhD of Economics 博士