Tell us what's eating your week
A short call, no slides. We'll tell you honestly whether software or AI is the answer.
A short call, no slides. We'll tell you honestly whether software or AI is the answer.
Services / 03
Language models fine-tuned on your experts' knowledge, running on servers you control.
Start a project →What we build
A general AI model knows a lot about the world and very little about your business. We take an open model such as Qwen or Llama, train it on your documents and expert answers, and run it on your own hardware or a private cloud. Your data stays with you, and the running cost is a server bill, not a charge for every question.
Starts at 120,000 THB, depending on complexity.
Hosting and maintenance are available on a monthly plan.
Private AI models
Answers questions the way your senior people would, from your contracts, manuals, protocols and past cases.
Sorts thousands of documents, tickets or transactions a day for a fixed server cost instead of a fee per message.
Writes quotes, reports and replies in your format and tone, ready for a person to check.
Patient notes, financial records and legal files stay on your own servers, which makes PDPA compliance simpler.
Trained on your own Thai and English material, so it uses your industry's terms correctly.
The same model can power your customer agents and internal tools.
Private AI models
We first test whether a hosted model with your documents is good enough. If it is, we'll tell you and you'll save the cost.
We gather and clean your documents and expert answers, and write a test set of real questions with the right answers.
We fine-tune the model and score it against the test set, side by side with the hosted alternatives.
We install it on your hardware or private cloud, monitor quality, and retrain as your knowledge changes.
Private AI models
| Hosted AI | Hosted AI with your documents | Fine-tuned private model | |
|---|---|---|---|
| Where your data goes | To the AI provider | To the AI provider, a piece at a time | Stays on your servers |
| Knows your business | No | Looks things up in your documents | Trained on it, uses your terms and style |
| Running cost | Per message | Per message | Fixed server cost |
| Set-up effort | Lowest | Moderate | Highest |
| Best for | General tasks | Information that changes often | Sensitive data, high volume, specialist knowledge |
Clinics, law and accounting firms, property groups, and any business with specialist knowledge or data that can't leave the building.
Private AI models
We've delivered fine-tuned models on Qwen and Llama, and choose the base model for each project by language, size and licence.
Not necessarily. Smaller models run on a single workstation-class machine; larger ones can run in a private cloud account in your name. We size it with you before you buy anything.
Less than most people expect. A few hundred to a few thousand good examples is often enough, and we help you build the set.
Searching documents (often called RAG) is better when information changes often. Fine-tuning is better for style, specialist language and classification. Many systems use both, and we test before recommending either.
You do: the trained model, the data and the code.