SCIENCE
AHLAME DOUZAL
Co-founder & CSO
- Associate professor at Université Grenoble Alpes for 25 years.
- 20+ years working on AI for temporal data.
- 60+ publications. 10 PhD theses supervised.
COMPANY
Deep science.Company-building experience.Control for consequential AI.
Xpdeep is model-level technology: it changes what an AI model itself returns, so the decisions of intelligent machines and consequential industrial systems can be read, bounded and evolved under control.
Native structural explainability is the foundation: engineers read the model's learned structure from within it. The three loops — design, production and controlled evolution — are its first product architecture, supporting model engineering, diagnosis and frugality as well as bounded authority under customer control.
It complements the customer's engineering and operating systems rather than replacing them.
A Grenoble research spin-out from the Laboratoire d’Informatique de Grenoble, CNRS and Université Grenoble Alpes.
NATIVE STRUCTURAL EXPLAINABILITY · INDUSTRIAL AI · CONTROLLED MODEL EVOLUTION
01 FOUNDERS
The research and company-building experience behind Xpdeep.
COMPANY BUILDING
STANISLAS CHESNAIS
Co-founder & CEO
- Built Netsize from zero to €100M net revenue across 23 countries before its sale to Gemalto.
02 FROM RESEARCH TO COMPANY
Built on deep science.
Not added around a black box.
More than two decades of research on temporal data led to the architecture now developed and industrialized by Xpdeep.
Founded in 2023 as a Grenoble research spin-out, Xpdeep brings decades of research on temporal learning into industrial model engineering and deployment.
STEP 1 · UNIVERSITY RESEARCH
University research on temporal machine learning.
STEP 2 · MODEL ARCHITECTURE
Development of Xpdeep's structural model architecture and company formation in 2023.
STEP 3 · INDUSTRIAL PROGRAMMES
Industrial programmes, deployment engineering and commercial development.
03 INDUSTRIAL EVIDENCE
What industrial programmes
have demonstrated.
EMBEDDED TARGET
8% vs 12% of ROM
On a European Tier-1 OEM’s own embedded target: 8% of ROM for the Xpdeep model vs 12% for its state-of-the-art model (virtual temperature sensors).
SOURCE: UNNAMED EUROPEAN TIER-1
105 → 19
Up to 82% fewer input variables; 105 reduced to 19 at equal or better accuracy.
Source: Independent programme evaluation.
UP TO 35×
Up to 35× faster inference.
Source: Selected validated results across Confiance.AI and client programmes.
12+
Enterprises on the Xpdeep platform.
Source: Company figure.
Figures come from different programmes.
See the technology →04 PROGRAMMES & COMPANY
French deep-tech.
Built for strategic industrial systems.
INDUSTRIAL VALIDATION
Confiance.AI and client programmes
Where the inference-speed and model-size results were validated.
PUBLIC RESEARCH & INNOVATION
EU Horizon Chips JU · France 2030 · ANR
Public programme support. Not customer endorsement.
COMPANY
- Xpdeep SAS, France
ORIGIN
- Spin-out of the Laboratoire d’Informatique de Grenoble (CNRS, Université Grenoble Alpes)
05 WHAT WE ARE WORKING TOWARD
Machines Will Decide.
We Decide How.
Xpdeep is building a European deep-tech company around a foundational model architecture for consequential industrial AI. Our ambition is to make structurally understandable and controllable learned models a practical foundation for future machines, equipment and industrial products.
- Decisions the model can carry inside customer-set conditions.
- Controlled model evolution between operating periods.
Responsibility
Customers retain responsibility for system engineering, validation and authorized decisions, including applicable safety, regulatory and certification requirements.
Beyond that: products, processes and business models built around decisions machines may carry. Transformation is an ambition to establish through application-specific evidence, not a delivered result.
ENGAGE
Build the infrastructure
behind the next generation
of intelligent machines.
WORK WITH US
Talk to Xpdeep →JOIN XPDEEP
Join Xpdeep →RESEARCH OR INDUSTRIAL PARTNERSHIP
Partner with us →