LLM fine-tuning services on your company data
A model that speaks your company’s language
We fine-tune an open language model on your documents and examples: it learns industry terms, writes in your format and handles requests more accurately. Training and data stay inside the perimeter; quality is measured on a test set of your tasks.
What we build
Not every task needs fine-tuning. If the model only has to know what is in your documents, search over a knowledge base (RAG) is enough and cheaper. Fine-tuning is for when the model must behave your way: understand narrow terminology, answer strictly in your document format, classify requests by your rules or write in your house style. We start with a check on the base model, build a dataset from your examples, fine-tune the model (usually with LoRA adapters, faster and cheaper than full training) and compare the result with the base model on the same tasks.
Industry terminology
Construction, healthcare, law, manufacturing: the model stops confusing the terms and abbreviations of your field.
Your format and style
Customer replies, reports, letters and conclusions in the structure and tone your company uses, without manual rewriting.
Classification and extraction
Requests, documents and emails sorted by your categories and fields - more accurately than a general model.
A compact model for the task
A small fine-tuned model often handles a narrow task as well as a large one and needs fewer GPUs.
What the work includes
Priced per task: depends on data volume and model size. Estimate within 1–2 days after the brief; GPUs are listed separately.
- A check on the base model and with RAG - so we don’t fine-tune where it isn’t needed
- Dataset collection and cleaning from your examples, labelling with your experts
- A test set of tasks and quality metrics
- Fine-tuning (LoRA or full) on your servers or rented GPUs
- Comparison with the base model and a quality report
- Deployment of the tuned model in your perimeter and a retraining plan
How we work
- 01
Requirements
We clarify which data must not leave: personal data, trade secrets, critical infrastructure. We record your security team’s requirements and the use case behind the project.
- 02
Perimeter architecture
Where models, the knowledge base and logs live, who can access what, and how the perimeter connects to your systems. The design is agreed with security before any hardware is bought.
- 03
Pilot inside the perimeter
4–8 weeks on a real task: model, documents, roles. Acceptance criteria - share of correct answers with a source, speed, hand-offs to people - are agreed upfront.
- 04
Fine-tuning and scaling
If a base model with document search is not enough, we fine-tune it on your data. Users, use cases and capacity grow with the load.
- 05
Support
Monitoring, model and knowledge base updates, incident handling. The perimeter and the code stay yours.
Questions
How is fine-tuning different from RAG?
RAG gives the model access to your documents: it finds the right passage and answers from it. Fine-tuning changes how the model itself behaves: terminology, format, style. Often you need both, but RAG comes first.
How much data is needed?
It depends on the task: for format and style a few hundred good examples can be enough, classification needs more. We estimate the volume from your data at the start.
Does data leave during training?
No. We train on your servers or on rented GPUs you approve. The dataset and the tuned model weights stay yours.
How do we know it got better?
Before training we build a test set of your tasks and run both the base and the tuned model on it. The go-live decision is made on numbers, not impressions.
The perimeter is live - next comes support
After launch we support the perimeter: model and knowledge base updates, monitoring, incidents and improvements under a contract or an SLA.
Other development areas
- Closed AI perimeterA perimeter design for models and data: roles, audit logs, network segments. Built around data protection law and your security rules - before any GPU purchase.
- Private LLM on your serversAn open language model inside your perimeter: a corporate AI chat over your documents, an API for your systems, roles and audit logs.
- GPU and infrastructureGPU server sizing for your model and load, site readiness and orchestration. Hardware for the task, not the top of the catalogue.