Starting research position F/M Training and using LLM as proof assistant assistants

Inria, the French national research institute for the digital sciences; France; Sur place; EUR 20 000 - 40 000; Plein temps

Résumé du poste

A leading research institute in France is seeking a postdoctoral researcher for the LLM4Code project. This position focuses on applying machine learning techniques for automated code generation involving formalized mathematics. Responsibilities include setting up software infrastructure, determining training methods, and documenting results. The ideal candidate will have experience in machine learning, basic knowledge of Lean and Rocq, and strong communication skills. An attractive salary starting from €3085 gross per month is offered.

Prestations

7 weeks annual leave

10 extra days off due to RTT

Partial reimbursement of public transport costs

Possibility of teleworking

Access to vocational training

Qualifications

• Interest in constructing formalized mathematical libraries and machine learning.
• Basic knowledge of Lean and Rocq is essential.
• Experience in training models and performing inference.

Responsabilités

• Set up the software infrastructure to conduct experiments.
• Determine training methods and implement extensions of Rocq or Lean.
• Document software and publish scientific papers.

Connaissances

Machine learning techniques

Communication skills

Autonomy

Expertise in LLM

Knowledge of Lean

Knowledge of Rocq

Formation

Experience in training models

Outils

Git

GitHub

Hugging Face

WandB

Description du poste

Inria, the French national research institute for the digital sciences

Organisation/Company Inria, the French national research institute for the digital sciences Research Field Computer science Researcher Profile Recognised Researcher (R2) Leading Researcher (R4) Established Researcher (R3) Country France Application Deadline 7 Dec 2025 - 00:00 (UTC) Type of Contract Temporary Job Status Full-time Hours Per Week 38.5 Offer Starting Date 1 Jan 2025 Is the job funded through the EU Research Framework Programme? Not funded by a EU programme Reference Number 2025-09524 Is the Job related to staff position within a Research Infrastructure? No

Offer Description

The postdoc will take place within the framework of the défi Inria LLM4Code, which objectives are to apply machine learning techniques to the automatic generation of code, and in particular for proofs assistants libraries, such as Coq libraries.

The research will be carried out in the LIP laboratory of ENS Lyon under the local supervision of Cyril Cohen, within the CASH team and the remote supervision of Guillaume Baudart,Marc Lelarge et Nicolas Tabareau. However, there will be frequent meetings, and regular in person meetings, with co-supervisors and participants from the LLM4Code project.

Travel expenses are covered within the limits of the scale in force.

Assignments

With the help of people from the projects and their collaborators, the recruited person will work on the automated translations from, to and between libraries of formalized mathematics using modern AI techniques, and in particular Large Language Models.

Such translations may require nontrivial alteration of the proof assistants at use, such as the creation of plugins, the use of meta-programming, or rephrasing libraries in ways that are better suited for machine learning.

It will also require to set up agent scaffolding, fine‑tuning, and reinforcement learning on existing models, and evaluate the relevance of each method.

The expected outcome of this work are research publications and software production.

For a better knowledge of the proposed research subject

The state of the art, bibliography and scientific references are fast evolving. We suggest to read, for example:

The recruited person will be mainly in connection with Cyril Cohen in Lyon, and Guillaume Baudart, Marc Lelarge, and Nicolas Tabareau, but will also work remotely with other members of the LLM4Code project.

Responsibilities

The person recruited is responsible for setting up the software infrastructure to conduct the experiments and take initiatives for the choices of software, and available hardware, within the limits allowed by Inria.

Main activities

• Determine how to train (finetuning and/or reinforcement learning), and use agent scaffolding on existing models, and combine these methods, to achieve the above mentioned assignments.
• Setup the infrastructure to implement and reproduce these methods.
• Implement extensions of Rocq or Lean in order to extract relevant data.
• Document the resulting software.
• Collect outputs, run benchmarks and publish scientific papers.

Additional activities

• Design user interfaces to make the resulting tool widely usable.
• Determine a long term maintainance scheme for these.
• Setting up scripts and processes for non-specialist users.
• Know the basics of at least Lean and Rocq.
• Understand the difference between various machine learning techniques.
• Have hands on practice in training models.

Languages :

• Advanced English.

Relational skills :

• Communicate about your results.
• Work autonomously, while talking about strategic decisions with other members of the project.
• Listen to feedback.

Other valued appreciated :

• Expertise in LLM.
• Expertise in Rocq or Lean.
• Git, GitHub, Hugging Face, WandB, and other platforms for code sharing, model sharing, etc.
• Runing code on Jean Zay

Specific Requirements

The candidate should demonstrate an interest in the construction of formalized mathematical libraries and machine learning. Although an in-depth training in proof assistants is not expected, it is important to know the basics of both Lean and Rocq. The candidate must have experience in training models and perform inference with them, and it is expected to have sufficient knowledge to quickly update oneself regularily.

Languages FRENCH Level Basic

Languages ENGLISH Level Good

Additional Information

• Partial reimbursement of public transport costs
• Leave: 7 weeks of annual leave + 10 extra days off due to RTT (statutory reduction in working hours) + possibility of exceptional leave (sick children, moving home, etc.)
• Possibility of teleworking (90 days / year) and flexible organization of working hours (except for intership)
• Social, cultural and sports events and activities
• Access to vocational training
• Social security coverage under conditions

From 3085 € gross salary / month

Selection process

Applications must be submitted online via the Inria website. Processing of applications submitted via other channels is not guaranteed.