STEP 01
Scoping
2 to 5 days · week 1
We read your data and your process as it is, we mock up, we settle what is feasible and we cost the rest in days. You leave with a plan and an estimate, even when the answer is “no AI here”.
AI engineering firm
Let’s build faster. Stronger.
We build AI into your processes: model, API, applications, infrastructure. We have already done it four times, end to end. The proof is below, screen by screen.
Our expertise
Four things we have already built.
One engineering foundation, four families. Each card opens on a proof: the real product, running, what it took to build, and what it delivers. Then on the form, pre-filled.
01
Automation & AI agents
“Three people retype orders received by email into the ERP.”
02
Document intelligence
“The answer is in our 40,000 documents; nobody can find it.”
03
Computer vision
“My inspection is visual, so it depends on who’s looking.”
- 10.1% relative error
- 1.72 cm median error
- n = 1,340 objects
04
Your AI product, from scratch
“We have the idea and the data. We’re missing everything else.”
How we work
Days of engineers who have already built all this, billed by the day, no opaque fixed price. A demo every week. Everything produced belongs to you.
WORK FOR HIRE · BY THE DAY
Why by the day? Because an AI project reveals itself as you build it: the data is never quite what you thought, the first model learns what the second will correct. A fixed price pushes you to hide that; day rates push you to show it. Every week you see what runs, you steer, you stop whenever you want, and what was built stays yours.
The partners
Two partners, and nobody between you and us.
No salesperson promising what they will not build, no junior learning on your project. The person who answers you is the person who builds.
Engineering and delivery
Models, API, applications, infrastructure. The four products on this page came out of here.
Business and the field
The first conversation, the scoping, and the industrial network in the Rhône-Alpes region.
SAS · Rhône-Alpes · all of France · no outside funding
The proof
A model in production, and its figures.
Estimating an object’s 3D dimensions from a single photograph. Study AlgoV6_1.1.5, evaluation of 6 July 2026.
0.0%
Relative dimensional error
0.00cm
Median error per object
0.0%
Objects more than 15 cm off
0
Validation objects
SCOPEThese measurements are established on a public validation set and ARKit, not on a client’s real inventory. They say what the model can do; yours would be measured on your data, with the same rigour.
What these figures demonstrate is not “we do vision”. It is that we know how to take a hard problem, train a model, evaluate it seriously and serve it in production. The same team built a renovation platform, a civic app, and this site.
Open
Tell us what costs you the most time.
Three minutes, four fields, and your workspace opens: your project, its timeline, and the thread with us, without waiting for our reply.
You writeWe readYour workspace opens
The path
How it goes, and what it costs.
By the day, from the first conversation to operations. No fixed price: a quote in days, reassessed every week with you.
STEP 02
Build
By the day · 4 to 16 weeks
Model, API, applications, infrastructure: a team delivering in one-week sprints, with a demo every Friday of what actually runs. You steer at every sprint; you stop whenever you want, and what was built stays yours.
STEP 03
Operations
By the day, on demand · as long as needed
Monitoring, drift detection, retraining on new data, evolutions. Or a full handover to your team: everything is documented for it.
Trust
Your data, your property, your compliance.
It is the first question any executive asks when offered AI. Here are the answers, written before you have to ask.
You own what is delivered
Models trained on your data, specific code, processed data: all of it belongs to you. We keep our foundation and grant you a perpetual licence on the delivered scope.
Your data stays yours
Deployment on your infrastructure or on a European cloud, self-hosted open models when the subject requires it. One client’s data is never used to train what is delivered to another.
Reversible, documented
What is delivered is documented and can be taken over by another team. No lock-in: it is a condition, not a selling point.
Compliant by design
GDPR from the design stage, risk classification under the European AI Act established during scoping. No vision on people or on public spaces.
The limits
What we don’t do.
A short list beats a page of values. It tells you in thirty seconds whether we are the right people to talk to.
No consulting without production. No master plan, no use-case mapping, no awareness seminar. We deliver software that runs.
No fixed price on a result nobody knows yet. An AI project reveals itself as you build it. We bill days, and we show every week what they produced.
No black box. Code, models, data, documentation: all of it is yours, and can be taken over by another team.
No vision on people. Nor on public spaces. It is a design choice, and it removes the regulatory risk before it arises.
No days without a demo. Every week, you see what runs. If nothing runs, we tell you first.