Data integration across company systems

Every system speaks differently: a different format, different names for the same thing, a different structure. Mapping into one model is most of the work – and without it nothing else can be built.

What if a source has no API?

We also handle exports, database dumps, e-mail attachments and, as a last resort, reading from the web or from images – a photo of a meter, a screen or a scanned document, for instance, which today works reasonably reliably. What matters is that access is stable and legal, not how it works.

Can you read data directly from machines?

We work with data that is already readable somewhere – in a database, an export or the API of the system that runs the machines. Connecting directly to machine controls is work for the machine supplier or its integrator; we agree with them on how the data is handed over.

We match using rules as well as similarity. Borderline cases are confirmed by a person, so we never blindly merge two of your customers into one.

An export gets a new column, an API changes its structure, a website stops responding. Validation on input and output catches it before an output someone relies on breaks.

For every value we record where it came from, when it was loaded and how it was calculated. Without that, nobody in the company will trust the data – and rightly so.

This is the most common reason we take on processing at all. You get the result as an API, an export to other systems or reports that generate themselves.

Do we have to change our systems for this?

No. Your systems stay as they are; we take the data from wherever it is already readable. If you also want a tool people work in – records, approvals or forms – that is a custom information system, and we connect it to the unified data.

Who on our side should be in charge?

One person who knows where your data comes from and one who decides what they want to see in it are enough. We take care of the technical part, talking to your system vendors about access, and running it.

When a rule changes, we replay the history – no manual digging. We build on proven open source, not a black box.

What technologies do you use?

PostgreSQL, Node.js, Python, Elasticsearch and Redis. Runs are scheduled to suit the environment – from cron to Airflow.

Without it, a model guesses with confidence and search returns three versions of the same thing. That is why data processing is often the first step before AI integration.

Will we then be able to ask the data questions in plain language?

Yes. On top of the unified data we build an assistant that answers with a link to the source, or a search that finds even what you cannot name precisely. It is the most common next step.

1. 1
   Source analysis We pick one question nobody can answer quickly today – a report put together by hand, for instance – and go through the sources it can be built from. That gives us the data model, scope and price.
2. 2
   Pipeline and checks Collection, cleaning, matching and validation – automatic, repeatable and with a record of every run.
3. 3
   Operation and development Monitoring of runs, reporting of anomalies and adding more sources or outputs as needed.

The first question and the first output show within weeks what working with the data is like, and you decide whether to continue. Further sources are then priced on what we actually find, not on an estimate.

Processing that runs every day needs someone watching over it. That can be us – but the process and rules are documented and run on your side.