Education
Tutors and assistants that teach with your method and your syllabus, without sending student data to the cloud.
We install a private AI at your school or college that knows your syllabus: tutors for students, help preparing materials and a search tool for your documentation, in Catalan and Spanish. Minors' data never leaves the school. We start with a free assessment at your Barcelona school or by video call.
An assistant that answers students' questions using your notes, examples and way of explaining, and cites the material each answer comes from.
Exercises, draft rubrics, adaptations for students with specific needs, and Catalan and Spanish versions based on your own content.
Datasets and LoRA fine-tuning of an open model so it can explain a system of your own that no general model knows, as we do with Nuzic.
Regulations, the school's educational project, curriculum plans and minutes, searchable with citations by teachers and the leadership team.
AI literacy for teachers and staff (Art. 4 of the EU AI Act), with classroom cases and clear privacy criteria.
A shared model server, with accounts per group, usage limits and logging, for universities, secondary schools and vocational training (FP) centres.
AI is already in the classroom, like it or not: students use it for homework and teachers to prepare lessons. The problem is where that data goes. In a school with minors, every assignment pasted into a cloud chat is a student’s information in the hands of a third party. In Barcelona and across Catalonia, teaching staff are already asking for local AI for exactly that reason, and the Catalan free software community is promoting self-hosted tools for schools.
Something similar is happening at universities, on a different scale. Campus AI centres are starting to buy their own model servers (the University of Valladolid has already put one out to tender), and many departments have GPUs that nobody has put to work.
Nuzic is an educational project from Barcelona that teaches music and mathematics with its own system. A general model does not know it and tends to answer with classical music theory. For Nuzic we built datasets from its materials, with step-by-step explanations, examples contrasting its system with classical theory, and conversations in both a technical and a classroom tone, and we LoRA fine-tuned an open model so the AI can explain its own musical and mathematical system. It is the same path any training centre or publisher with its own method can follow.
A vocational training student asks at eleven at night how a sales margin is calculated, and the tutor answers with the example from his teacher’s notes, in Catalan. The teacher prepares three versions of an exercise for different levels in ten minutes. The leadership team asks what the fourth-year curriculum plan says about project assessment and gets the exact paragraph. We have already built a tutor like this for a vocational training student, with the official curriculum for the course and voice support.
Catalan schools work in Catalan, and an AI that writes it badly is no use in class. We follow the AINA project and Softcatalà closely, and we have built a Catalan vocabulary pipeline that raised the coverage of the ‘draft’ model that speeds up responses from 75.2% to 99.6%, so that Catalan is not a second-class language.
If for now you only want teachers to learn to use AI sensibly, start with training: the server can wait.
AI and GDPR assessment
We review how you use AI today, what data it touches and what obligations you have. You leave with a clear plan: what to run locally, what can go to the cloud and in what order.
Ready-to-use local AI
Your own ChatGPT, inside your office. We choose or reuse the hardware, install the models and set up a private chat for the whole team.
Private RAG search
Ask your documents as you would ask an expert colleague. The AI answers citing the document and page, and respects who can see what.
Custom LoRA model
A model that speaks like your business: your jargon, your formats and your tasks. We build the dataset, train the model and measure whether it really improves.
AI training
A hands-on half day so the team uses AI well: real cases from your work, clear rules on data and the AI literacy the EU AI Act requires.
Yes, if the model runs on a machine at the school itself. That way, students' work, questions and names do not go out to third-party services, and the school decides who has access and what is stored.
It means training an open model a little further with your examples so it learns your method and your tone. It makes sense when you teach a system of your own that no model knows; if you just want it to answer from your notes, search with citations (RAG) is enough and cheaper.
Yes, in many cases. We check which models fit in its memory, set it up with accounts per group and measure real speed before promising anything; if you need something more, we help you choose it. We look at this in the free assessment, at your Barcelona school or by video call.
Yes. The models we install work in Catalan and Spanish, and we have expanded their Catalan vocabulary so that answering in Catalan is no slower. In the pilot we test it with your real materials.
We install a private AI at your school or college that knows your syllabus: tutors for students, help preparing materials and a search tool for your documentation, in Catalan and Spanish. Minors' data never leaves the school. We start with a free assessment at your Barcelona school or by video call.