AI you can actually trust, 100%

We combine the creativity of generative AI with the reliability of mathematical formal methods. By leveraging the verification capabilities of formal methods, such as the Lean 4 theorem prover, we pioneer a new generation of Large Language Models (LLMs) that are reliable, transparent, and efficient.

Our Value Proposition

We provide four unique formal guarantees to our LLMs.

100% Logical Consistency reliability

Zero reasoning hallucinations, even on arbitrarily long chains of thought.

100% Enforceable Constraint reliability

Add your regulatory, legal, or business rules once, and it becomes formally impossible for the model to disregard them.

100% Traceability transparency

Every decision is accompanied by a fully transparent proof, ending the “black-box” era.

Significantly Lower Cost efficiency

No more expensive guardrails, human-in-the-loop reviews, LLM-as-a-judge pipelines, or over-sized models to compensate for unreliability.

Use Cases

Our solution applies to three critical business needs.

Trustworthy Generative AI (LLM & Vision-Language Models)

100% logical consistency, 100% enforceable constraint, 100% traceability, and significantly lower cost.

A Robust Alternative to Traditional Machine Learning

Statistical models give you probabilities. We give you certainty. Our engine directly executes and verifies your exact business rules as formal specifications — no training data drift, no false positives/negatives, no retraining when rules evolve.

Document & Decision Logic Auditing at Scale

Instantly prove consistency across thousands of documents, contracts, or policies (single- or multi-source). Retroactively audit historical decisions (e.g., “Were these 10,000 insurance claims or loan approvals processed according to the rules in force at the time?”). Guarantee your entire knowledge base is logically coherent and regulation-compliant.

The Team

The two co-founders met during their Ph.D. at ENS Paris-Saclay.

Sylvain Combettes, Ph.D.

CEO

Sylvain was a Senior ML Product Engineer at Probabl, the startup spin-off from Inria and official operator of scikit-learn. He holds a Ph.D. from ENS Paris-Saclay. He was a data science lecturer at École polytechnique EXED, École polytechnique (X-HEC Data Science for Business MScT), and CentraleSupélec EXED.

Antoine Mazarguil, Ph.D.

CTO

Antoine worked at Qubit Pharmaceuticals, first as a software engineer working on the company's HPC calculation management, then as a ML engineer. Antoine holds a Ph.D. from ENS Paris-Saclay and graduated from both École polytechnique and MVA (Mathématiques, Vision, Apprentissage).