AI solutions that pay for themselves.

We put AI to work where it saves real effort: assistants that answer from your own documents, automatic processing of invoices and contracts, and AI features inside your software. We start by working out what it will save and what it will cost.

Start an AI projectSee how we work
Example: AI reads an incoming invoice and extracts six fields, each with a confidence score. The five certain fields are recorded in accounting, and the one uncertain field goes to a person for review. Separately, an AI assistant answers a customer question and cites its source in the documents.

ExampleAn incoming invoice, into accounting

Extracted data

Supplier
Example Trading Ltd
confidence 99%
Company ID
2034•••••
confidence 99%
Date
6 Oct 2026
confidence 98%
Net amount
€1,240.00
confidence 97%
VAT 20%
€248.00
confidence 97%
IBAN
BG•• •••• 4417
confidence 74%. To review
  • 5 fields recorded
  • 1 field waiting for a person

AI assistant

What warranty do the air conditioners have?

Five years on the compressor and two years on all other parts.

Source: Warranty terms, section 3.1

When AI pays off, and when it doesn't

An AI solution is software that uses language models (like the ones behind ChatGPT) or computer vision to do work that used to need a person: reading documents, answering questions, sorting and checking data.

It pays off when a task repeats hundreds of times a month, when the information is scattered across documents and emails, and when a mistake can be caught by a check. It doesn't when the rules are simple and clear: then ordinary automation is cheaper, faster and more predictable. We'll tell you honestly which of the two you need.

What we build with AI

The four kinds of AI solution that most often bring real value.

  1. 01

    An AI assistant that knows your documents

    A chatbot for your website or an internal assistant for your team that answers from your policies, catalogues and contracts, and cites the source of every answer. When it doesn't know, it says so and hands over to a person.

  2. 02

    Automatic document processing

    Invoices, waybills, contracts and orders are read automatically: AI extracts the fields you need, checks them and records them in your accounting or inventory software. Uncertain cases go to a person.

  3. 03

    AI features in your product

    Search that understands what a question means, summaries of long texts, automatic categorisation and recommendations, in the app you already have or the one we build together.

  4. 04

    Computer vision

    Object recognition, quality checks and reading signs and meters from a camera, including right on the device, so images never leave your premises.

How we bring in AI, step by step

We start small, measure, and scale only what works.

  1. 01

    Pick the task

    One task that repeats often and is easy to measure: for example, how long it takes to enter one invoice.

  2. 02

    Collect real examples

    Dozens to hundreds of your real documents or questions, with the right answers. They become the benchmark we measure against.

  3. 03

    A prototype on your data

    A working prototype in a few weeks. You see the accuracy in numbers, not in a slide deck.

  4. 04

    Roll out with human review

    We go live gradually: at first a person approves every result, then only the uncertain ones.

  5. 05

    Monitor and improve

    We track accuracy and cost every month and switch models when a better or cheaper one appears.

What you get

  • A working solution connected to your systems
  • Accuracy measured on your own real examples
  • Cost control for every request to the model
  • Clear rules on which data is processed where
  • Documentation and training for your team
  • Support and model updates
See the full process

Technology and privacy

We choose the model by the task, the price and where your data has to stay, not by fashion.

Language models
  • OpenAI
  • Anthropic Claude
  • Google Gemini
  • Mistral
  • Llama

Models from the major providers, or open models on your own server when data must not leave the company.

Document search
  • RAG
  • pgvector
  • PostgreSQL

Documents are split into passages and indexed by meaning, so the assistant finds exactly the right one.

Computer vision
  • PyTorch
  • ONNX
  • On-device models

We train and optimise models that also run on small devices, without the cloud.

Integration
  • Python
  • TypeScript
  • .NET
  • REST API

AI features connect to the software you already use through documented interfaces.

Privacy
  • GDPR
  • EU data centres
  • DPA

Your data is not used to train anyone else's models. We work with providers that have EU data centres and sign a data processing agreement (DPA).

Questions about AI solutions

How much does an AI solution cost?

It depends on the task and on how many systems it connects to. A pilot for a single task usually costs a few thousand euros. On top of that comes the monthly cost of requests to the AI model, which depends on volume. Before we start, we calculate both for your case, along with the hours the solution will save.

Is our data safe?

Yes, when the solution is designed properly. We work with providers that have EU data centres and a data processing agreement, and that don't use your data to train their models. When data must not leave the company, the model runs on your own server. Access to documents follows the permissions your people already have.

Can AI make mistakes or invent an answer?

It can, which is why we design solutions so that mistakes get caught. The assistant answers only from your documents and cites the source, and when the information isn't there, it says it doesn't know. For documents, every field has a confidence score and uncertain cases go to review. We measure accuracy on your real examples before launch.

How long does it take?

A working prototype on your data is usually ready in a few weeks. Putting it into everyday work usually takes one to three months, depending on the data and the systems it connects to.

Which AI models do you work with?

OpenAI, Anthropic and Google models, and open models such as Llama and Mistral that can run on your own server. The choice depends on accuracy for the specific task, on price and on your data requirements. Our solutions aren't tied to a single provider.

Is AI worth it for a small business?

Yes, if you have a task that repeats often: answering the same questions, entering documents, sorting enquiries. If ordinary automation will do the same job for less, we'll recommend that instead.

Related services

Which task should AI take on?

Start an AI project

Prefer to talk?

Talk to an engineer, not a salesperson.

+359 884 139 349Mon-Fri, 9:00-18:00
CallStart a project