Service

AI integration and business automation

Practical AI added to real processes, with the rules that keep it reliable. The assistant on this website is a working example.

Where AI helps, and where it doesn't

AI is good at reading, summarising and understanding language. It is bad at being exact without supervision. So we use the model for language and leave numbers, decisions and permissions to ordinary, tested code.

What we build

• Assistants for your website or team, answering from your own documents.
• Lead qualification that gathers the details you need and passes on a structured brief.
• Document processing: reading documents and extracting data into your systems.
• Process automation combining AI steps with normal integrations.
• Integration of language models into your existing product.

Example: the assistant on this website

The project assistant on our homepage talks to visitors and gives an indicative price. It is built so the model cannot make up a number:

• The model gathers requirements in conversation. It never sees the price list.
• A separate pricing engine calculates every amount from a catalogue, and projects above a set threshold go to a personal consultation with no automated figure.
• Output guards reject any model reply containing an amount, rate or duration, and fall back to a scripted question.
• It runs on Gemini through Vertex AI in the EU, and it has automated tests including the critical price rule.

Example: self-hosted face detection in Kindora

In Kindora, a self-hosted face detection model (UltraFace on ONNX Runtime) blurs children without consent in photos. Nothing is sent to an external AI service, and if the detector is unavailable the pipeline fails closed: the photo is not published.

Technologies

Gemini on Vertex AI, Azure OpenAI and the OpenAI API, self-hosted ONNX models, with .NET or Node.js backends. We pick the model and the hosting region according to your data and requirements.

Questions and answers

Is our data used to train AI models?

We choose providers and settings where your content isn't used for training, and we explain the data flow before implementation. Variable AI usage costs are billed separately from the project.

What will it cost?

Describe your idea to the assistant for an indicative range; the final price and timeline are agreed after an individual review.

Related

• Custom software development: Business applications, internal systems, APIs and integrations built around how your company works.
• Kindora: nursery and kindergarten software, with a privacy pipeline for children's photos.