Clémence Lanfranchi

Clémence Lanfranchi is a researcher at Mistral AI whose work spans graph learning, language models, and reasoning systems. At ETH Zurich, she completed a semester project on suspicious-registration detection using graph neural networks. In 2022, she co-authored a study of e-commerce fraud on dynamic graphs. In September 2024, she was one of four Oracle Labs speakers at an EPFL tech talk and was named on an Oracle patent application for LLM-generated table descriptions. Her 2025 papers include Distill-C for natural-language-to-SQL tasks and Magistral, Mistral's first reasoning model.

Now at: Mistral AI
Works: 11
Citations: 17
h-index: 3

Publication numbers from OpenAlex, as of 19 October 2026.

Timeline

2022

Fraud graphs with attention

Her ETH Zurich semester project explores temporal GNN methods for suspicious mass-registration detection with Shuai Zhang and Ce Zhang as joint supervisors. In 2022, she co-authors a study of e-commerce fraud that models dynamic heterogeneous graphs with attention.
September 2024

An Oracle Labs talk

In September 2024, she is one of four Oracle Labs speakers at a tech talk hosted at EPFL.
2024

An Oracle LLM patent

In 2024, an Oracle patent application names her and two collaborators among its inventors for a method that uses a large language model to generate natural-language descriptions of data tables.
2025

Customizing models for SQL

In 2025, she co-authors Distill-C, a framework that uses synthetic data from teacher LLMs to fine-tune smaller open-source models for natural-language-to-SQL tasks.
2025

Co-authoring Magistral

In 2025, she co-authors the paper introducing Magistral, Mistral's first reasoning model, alongside a scalable reinforcement-learning pipeline.
Since 2024

A researcher at Mistral

In 2026, she co-authors Ministral 3, a family of parameter-efficient dense language models; she is a researcher at Mistral AI.