Large language model (LLM) fine-tuning
Fine-tuning specialises an open-source model in your field: your vocabulary, your style, your tasks. The result is a model you own and can host wherever you like.
Our fine-tuning process
Choosing the base model
An open-source model whose licence allows commercial use and modification.
Preparing the data
High-quality examples, built and checked with your experts.
Training and evaluation
GPU training, then tests on real cases before going live.
Deployment
The model runs on your machines or with a European host.
Real-world examples
- A legal model trained on French and Italian law for LegisBox.
- A model that writes technical reports in a company’s exact format.
- A model that sorts customer requests using an internal taxonomy.
Frequently asked questions
Fine-tuning or a RAG assistant: which one?
RAG brings up-to-date knowledge from your documents. Fine-tuning teaches the model a way of reasoning, writing or classifying. They are often combined.
How many examples are needed?
A few hundred carefully built examples are often enough for a specific task. Quality matters more than quantity.
Related services
AI app development
We build business software that uses AI to read, sort, summarise or draft for you.
AI agents and automation
An AI agent chains several actions for you: it reads an email, updates the CRM, prepares an invoice and schedules a follow-up.
AI assistants on your data
A RAG assistant answers your team’s questions using only your documents and shows where each answer comes from.
Dataset creation
A model is only as good as its data.
Let’s talk about your project
A free first conversation, with no commitment, to see what would really help your business.