The company behind the systems
Gwenlake designs, develops and operates data and AI systems for companies that need them to work in production. We are a team of data scientists and engineers who care as much about how a system runs as about how it performs.
Vision
We believe in a positive technological future. Not an inevitable one: AI will make expertise cheaper to reach, in medicine, in industry, in the services people depend on, only if the systems carrying it are built to be relied on. That is the work.
We believe artificial intelligence will do most of its useful work inside organisations, on their own data, under their own control. Not in a chat window, and not on someone else’s servers.
We believe most AI projects die between the data and the model. Not for lack of models: for lack of connected, governed, well-described data, and for lack of engineering around the model once it runs. That gap is where we work.
We believe a reliable AI system is built on four things: data with provenance, an ontology that says what the data means, evaluation on real cases, and an environment the organisation controls. Everything we build follows from that.
How we got here
Gwenlake was founded in Rennes in 2021 by a small team that had one foot in the university and the other in production systems. Our first projects were about scientific text in the medical field: extracting, classifying and tagging literature at scale. They taught us something we have kept ever since: a model is only as good as the domain expertise and the data around it.
We then moved from text to signals, analysing sensor data from ships at sea. Noisy data, large volumes, and a real cost when the answer is wrong. That work forced us to take the engineering as seriously as the models: pipelines, provenance, evaluation, and what happens on the day the system fails.
After the sea came signals of another kind: brain data. Recordings that are shorter, noisier still, and where the ground truth is itself open to interpretation. It pushed us further in the same direction, putting less weight on the choice of model and more on how the data is described and how a result is checked.
Each of those projects left something behind. Running models privately for clients became the Gwenlake Inference API. Chaining models into systems that act, with tools, memory and a check on the result, became Gwenflow, the open-source framework our agents run on. And the platform we kept rebuilding from one project to the next, to connect a client’s systems, prepare their data and write results back, became Gwenlake Atlas.
Today, our job is to make AI work in production for the organisations we serve: pipelines, models and agents, designed, built and operated by one team, on Atlas. What has not changed: part of the team still teaches and does research, in the same city we started in.
Core values
Curiosity
Last year’s answer is rarely this year’s.
Commitment
A system in production is a promise, not a delivery.
High standards
The work is finished when it is right, not when it is due.
Humility
The day you stop doubting your own model is the day it fails somebody.
Client first
A project is done when it works for the client, not when it ships.
One team
The same people scope the problem, train the models and ship the application.
Ship and run
We ship quickly, and we operate what we ship.
No silos
No heroes and no silos: the team works as one.
Small and proven
Small, well-evaluated models beat large, unevaluated ones in production.
Upstream wins
Data quality is worth more than model choice. Most of the gain is upstream.
Find the error
Lineage is not compliance: it is how you find the error the day the answer is wrong.
Explain every run
Every run should be traced. If nobody can explain why the system did what it did, it is not done.
Sovereign by design
Sovereignty is an architecture decision, not a feature. It is decided on day one or never.
Our academic roots
We never fully left the university. Part of the team still teaches at the University of Rennes, and we remain active in a research group in the same city we are based in. Every year we also organise an international quantitative finance conference, bringing researchers and practitioners together around the kind of open exchange we try to build our products with.