AI agents & assistants

AI that does real work — with a person in charge.

Agents and assistants that read documents, answer from your own knowledge and draft the routine — inside your systems, with every important decision approved by a person.

Situations

Where AI earns its keep

Not chatbots for the sake of it. Jobs that take people hours and follow a pattern.

Long documents, read in minutes

Tender books, contracts or regulations summarised, with the requirements and risks pulled out.

ExampleLicitor: AI analysis of tender documentation

Search by meaning, not just by words

Find the right record even when people describe it differently than it was written.

Examplesemantic search in Licitor

Email sorted and drafted

Incoming messages classified, routine ones answered as drafts for a person to approve.

An assistant over your procedures

New colleagues ask the assistant instead of interrupting the senior ones — answers cite the source.

Data pulled out of PDFs and invoices

Fields extracted from documents into your systems, with a person checking anything uncertain.

Reports and assessments at scale

Consistent written assessments generated from measured data, reviewed before they go out.

ExamplePrestigiu: website audits on 8 criteria

What we work with

The right model for each job.

We pick the model by task — a fast, cheap one for sorting, a strong one for analysis — and track what every request costs.

  • Claude API
  • embeddings
  • pgvector
  • agents with tools
  • human review
  • cost tracking

When it is not worth it

When a simple rule would do.

If a plain rule solves the problem, we do not use AI: it is cheaper, faster and fully predictable. We use AI where rules break down.

In production

Running today

A few of the systems where this already works, every day.

Questions

Before you ask.

Is our data used to train AI models?

We use business APIs whose terms exclude training on your requests, and we only send the data a task needs.

What if the AI gets it wrong?

That is why a person approves anything important, and why we test on your real cases before launch.

How much does it cost to run?

It depends on volume and model. We estimate it before we start and track the real cost per request afterwards.

Next step

Which job would you give an AI first?