AI and automation

Start with the value

Many AI projects fail for a simple reason. Nobody checked whether the task was worth automating.

We begin with a short review. It ranks the possible uses in your operation by benefit and by difficulty, and states clearly which ones to drop.

What we cover

• Use-case review and ranking, with expected benefit and effort
• Custom model development, trained on your own data
• Data labelling and annotation, with the team to carry it out
• Fine-tuning of open models for your domain and your languages
• Training and inference infrastructure, on cloud or on hardware you own
• Document and data automation: extraction, classification, reconciliation, reporting
• Internal assistants built on your own documents and systems
• Process automation between the tools you already use
• Evaluation, monitoring and retraining as your data shifts
• AI governance: data protection, retention, rules of use for staff
• Team training, from management awareness to daily practical use

Building and training models

Not every problem is solved by calling an external service. When a task is specific, repetitive and based on your own data, a smaller model trained for it is often cheaper to run and easier to keep under control.

We handle the whole chain: preparing and labelling the data, choosing the architecture or the open model to start from, fine-tuning it, evaluating against a test set you agree on, and putting it into production. Labelling teams can be recruited and managed locally.

The languages of the region

General-purpose services handle Khmer, Lao and, to a lesser degree, Vietnamese poorly. A model fine-tuned on your own material performs better on these languages than a larger generic one, and it stays on infrastructure you control.

Infrastructure

Training and inference have different needs. We size both, on cloud GPUs or on hardware you own, and we keep the running cost visible instead of letting you discover it on an invoice.

Your data, and where it goes

Sending company documents to an external model has legal and organisational consequences. This matters for organisations that report to donors or to a European head office.

We treat it as a governance question first. Sometimes the right answer is to keep a workload entirely off external services.

What we will advise against

A chatbot that answers worse than your staff. Automating a process that should be removed instead. Replacing a person whose real work is judgement. The review exists to identify these early.