AI consulting
Hands-on workshops, proofs of concept and agentic workflows for companies navigating the AI revolution. The studio shows what to automate and how to keep humans in the loop, and builds with the methods it uses in its own work.
One request to an agent, from the question to the log: which deliveries are late this week, and who needs to be told?
- Orders
- Stock
- Emails
- Documents
- Approval by a person
Focus
- See the opportunity: what to automate, and how to keep humans in the loop.
- Build proofs of concept with classical machine learning, deep learning or LLMs, where each makes sense.
- Set up agentic workflows, in the cloud or on premise.
- Train teams hands-on, through workshops.
When to bring us in
A company wants to use AI and needs to know where to start and what it is worth.
Leadership has to decide on AI
The board asks where AI pays off and what it puts at risk. A proof of concept on one real case gives it measured results to decide on.
Teams want to learn by doing
People have read about agents and want to run them on their own tasks, without exposing company data.
Automation needs a person in the loop
A task can be automated, but a wrong action is expensive. The workflow needs approvals for the actions that matter and a log of every step.
What gets delivered
Each one can be booked on its own.
- Opportunity mapA ranked list of what to automate in your company. Each case comes with what it is worth, what could go wrong and where a person stays in the loop.
- Hands-on workshopA half day for a leadership or innovation team. People run agents themselves on a practice company, so nobody connects company systems or shares confidential data.
- Proof of conceptOne real case, built with your IT team, with the method that fits the problem: classical machine learning, deep learning or an LLM. Leadership gets results measured on your own examples and a recommendation to build, change or stop.
- Agentic workflowThe workflow running in the cloud or on premise, with approvals for the actions that matter and a log of every step.
Experience
The studio runs its own projects on the methods it teaches.
Practised in the studio
The studio uses agentic workflows in its own projects. This website has its own tools that AI agents can call.
In AI since 2011
Dr. Kirell Benzi, who leads the studio, has built AI systems since 2011, from classical machine learning to deep learning, LLMs and agents. He holds a PhD from EPFL.
Applied research
The studio turns AI research into working systems with the VITA lab at EPFL, whose papers appear at the field's top conferences.
Big data
Swiss Echoes makes 200,000 hours of Swiss television archives explorable. REMIT works on more than 120,000 hours of RTS video.
Many kinds of systems
Agents, retrieval with sources, semantic search, speech recognition, knowledge graphs, and generative AI for images and video.
What stays under your control
These rules apply to every project.
Your data
It stays yours. The studio has no rights to it, never uses it to train models and signs NDAs.
Where it runs
In the cloud or on your own servers, as you decide.
Humans in the loop
You decide which actions an agent takes alone and which wait for a person. The log records both.
Questions people ask
What is an AI proof of concept?
A small working version of one use case, built on your own examples, with measured results. It tells leadership whether to build, change or stop before a larger budget is committed.
What is an agentic workflow?
A business task handed to one or more AI agents, together with the rules around them: what they may read, which tools they may use, when they run, which actions wait for a person's approval and what goes into the log.
Can the system run on premise?
Yes. Agentic workflows can run with a cloud provider or on your own servers, when the data has to stay inside the company.
When is classical machine learning a better choice than an LLM?
When the task is a prediction on structured data, such as a forecast, a score or an anomaly. It costs less to run and is easier to check. An LLM fits tasks that need language or reasoning over documents.
Do you need access to our systems for a workshop?
No. A workshop runs on a practice company, so nobody connects company systems or shares confidential data. A proof of concept needs access, and your IT team decides what the agent may reach.
Where do you work?
Anywhere. The studio works with companies in any country, in French or English.
How much does AI consulting cost?
Each engagement is quoted from the brief.
Related work
A talk on AI agents for business leaders, and two AI projects with RTS and EPFL.