Services
Our services answer three questions.
They are the questions you face when a data or AI initiative is about to start, is under way or has stalled. You can start with any of them. Few organisations start from scratch, which is why we first ask what you have already done.
01
What to build, and in what order?
First you decide what needs to change in how you work. Only then, what to build.
02
How do we get a data foundation we can trust?
Trustworthy data comes from two things: the platform it is produced on, and the way the work is done.
Data platform
A platform in the public cloud, in your own environment, or both. What a production-ready platform requires.
Data platformData as Software
The method that makes data production testable and repeatable. The four changes that Western Uusimaa Wellbeing Services County made.
Data as SoftwareHealth data: standards and regulation
OMOP for research, FHIR for data exchange, regulation built into the structure.
Health data: standards and regulationHealth and social care situation centres and preparedness
A data foundation that serves management on a Monday and during a disruption.
Health and social care situation centres and preparedness03
How does it end up in use, and stay there?
A working model or a finished platform is not the same as one that is in use. Production use has six prerequisites, and they are needed at the same time.
AI into production
Six prerequisites, a model registry, a reference architecture and the AI Act timetable.
AI into productionTechnical ownership of the whole
In a multi-supplier project, someone has to be responsible for architecture, interfaces and quality requirements. We can take on that role for the duration of the implementation.
Technical ownership of the wholeHow we work
There are three steps, and each ends in a result that can be assessed.
Definition
We establish the problem and the change you are aiming for: who does the work today and what the change requires of data and AI. The result is a written plan for the next step, or a reasoned recommendation not to take it.
Limited verification in the real environment
We build the core of the solution and use it to verify the change: real data, real environment, real users, limited scope. The criteria for production use are written before the verification begins. That way you know whether the idea works before more money is committed to it.
Production and getting it into use
We take the solution into production, set up monitoring and do what is needed for the new way to replace the old one in professionals' everyday work. The work is finished only when the tools are in daily use by your team.
What you buy from us
You get experienced specialists and an agreed outcome. A large initiative can be done in several parts.
An agreed outcome
Project
An experienced specialist answers a defined question, and the result is agreed in advance. A definition, for example, or limited verification in the real environment.
A named specialist in your team
Continuous allocation
An architect, a lead developer or the person responsible for changing how the work is done fills a role in your organisation for as long as the work needs. Building a data platform, for example.
An implementation team
Overall responsibility
A typical implementation team has an architect, a lead data engineer and a data engineer. This is how our method is put into use in your organisation. The whole data production area, for example.
Integrator service
Technical ownership of the whole
We steer the implementation and represent your interests. We are responsible for architecture, interfaces and quality requirements, and we review every change before it goes into production. Your own team or several suppliers can do the build, and each supplier keeps contractual responsibility for its own delivery.
Questions and short answers
How do we modernise our data platform without locking ourselves into one supplier?
By deciding first what has to be portable: the data, the data models and the key workflows. Those are built on open standards and replaceable components. Beyond that, you can choose a single supplier's product, as long as the choice is a conscious one.
Data platformWhy do our data pipelines break every year?
Usually because the transformations are long SQL queries that cannot be meaningfully unit-tested, and the same logic has been copied to many places. Almost every fix then breaks something else. The remedy is in the method: small testable functions, version control, and automated testing and deployment.
Data as SoftwareHow do we get OMOP to work in practice?
OMOP provides the structure, but harmonisation is always an interpretation of what the source systems contain. What matters is how the interpretations are documented and who maintains them. Vocabulary mappings need version control just as code does.
Health data: standards and regulationOur pilot works, but progress has stalled. What is missing?
Most often an owner and criteria. Nobody has been named as the owner of the phase after the pilot, and the conditions for going ahead or stopping have not been written down. The model is rarely the problem.
AI into productionWe need a situational picture that management can steer by. Where do we start?
With the decisions. First define what is decided on the basis of the situational picture, and who decides. The data content comes next, and the views only at the end.
Health and social care situation centres and preparednessA tender is coming up. How do we know what to buy, and in what parts?
In a definition, we write the procurement requirements and acceptance criteria so that they are technology-neutral yet precise enough. At the same time you decide what stays under your own control.
DefinitionThere are many suppliers. Who takes technical responsibility for the whole?
In a multi-supplier project, one party has to be responsible for architecture, interfaces and quality requirements. We can take on that role for the duration of the implementation. We then review changes before they go to production, including those made by other suppliers. We have held this role with two customers.
Technical ownership of the whole