Custom
AI solutions
built around your process

A ready-made chatbot on your own knowledge base and a custom solution wired into your systems are tasks of a different order, so we quote after the diagnostic call rather than from a price list. What moves the number: the volume and state of the data we build the knowledge base from, how many systems we integrate with, where the model runs (a cloud API or your own servers), and how complex the logic is.

It depends on the scenario, the number of integrations and how ready your data is, so we name the schedule after the diagnostic. Wiring in an existing product takes days. A custom scenario starts at a week. A project integrated into your systems takes two to three weeks and up; a project with models deployed inside your perimeter starts at one to two months. In some cases we build a free prototype on a slice of your data before the project starts.

Doing it yourself needs an ML team, prepared data and months of experiments. Going through an integrator is faster: you bring the task and the data, we own the model, the integration and the launch, and your team keeps running the business meanwhile.

A task, data and access. The task is a specific process plus the metric that tells us it worked: response time, share of resolved tickets, number of errors. The data you almost certainly already have: policies, documents, conversation history. Access is needed to the systems the solution exchanges data with. We pick the model ourselves and confirm the choice with a prototype on your data.

No, that is the normal starting point. Loose policies, exports and correspondence are already material: we assemble a knowledge base from them and shape it into something a model can work with. At the diagnostic we say plainly what is missing and what has to be collected by hand. That is part of the work, not a reason to postpone the project.

Yes, though it usually costs more: a self-hosted model needs a GPU server, rented or bought. So while the task involves no sensitive data, most teams start on cloud LLMs. When the data is sensitive, we deploy an open-source model inside your perimeter: the model, the knowledge base and the logs never leave the company, which is what GDPR and internal security reviews ask for. We do it end to end, including the servers, the deployment and the access rules.

A model does not look up a ready answer, it completes the most probable one, so mistakes are possible. We have been shipping LLM solutions since 2023 and we know how to get the error rate down to a level the business accepts: choosing the right LLM and embedding model, preprocessing the data properly, tuning the agents, multi-agent architecture, hard validation in code and human review where the cost of an error is high. The mix is decided per project.

Sometimes yes: for standard tasks we will offer a ready solution, and a chatbot on your knowledge base deploys quickly. More often, though, buying AI means commissioning development around your own process. The value comes from the integrations with your systems and data; the model itself is only a part of the solution.

Business processes change and the solution grows with them: we refresh the knowledge base as your policies and products change, extend the scenarios to neighbouring tasks, and move to stronger models as they ship. You can run this yourself or hand it to us as support.