We design, deploy and secure AI systems, in the cloud or hosted on your premises.
Founded in 2026 and based in Toulouse, France, Hexo Signal has its roots in scientific research in applied deep learning, numerical simulation and industry. Our path, between research and industry, brought us face to face with the realities of the field: data from sensors and numerical simulations, production constraints, reliability requirements and the confidentiality of industrial data.
From this experience came our conviction: to build tailor-made AI systems that solve complex problems and bring real value to our clients’ businesses. We are also committed to developing AI solutions that are more local, more frugal and sovereign: systems deployed on specialised, secure infrastructure, sized to actual needs. It is an additional technical challenge, one that gives our clients full control over their data, predictable costs and greater technological independence.
Today we put this expertise to work for companies and institutions, across three key areas of the AI value chain:
The term “artificial intelligence” is everywhere today, in everyday language and in society at large. In reality, AI is a very broad discipline, made up of several distinct fields.
Algorithms that learn from data, instead of being programmed by hand, rule by rule
Machine learning
Deep learning
Generative AI
Our scope
A sub-branch of machine learning that uses model architectures based on neural networks.
Images
Quality control, defect detection
Time series and signals
Predictive maintenance, anomaly detection
Multimodal data
Digital twins, process monitoring
Numerical simulation
Reduced-order models, inverse problems
Non-exhaustive list
The branch of deep learning that produces text, code or images, and which includes LLMs and agents.
Non-exhaustive list
At Hexo Signal, we share the values of free software and open science, which help develop technologies that serve the common good. We build our AI solutions almost exclusively with free and open-source technologies, and we want to give back to this ecosystem.
In this spirit, we are developing a simulator for sizing LLM inference systems, from workstation to server (performance, energy consumption and cost), which we plan to release as open source and make available to the community.