Skip to main content

Dev Station Technology

Data engineering

The plumbing that has to work before any of the clever stuff

Ingestion, storage, movement and migration. We have run readings from device fleets into time series storage, moved years of sensitive records with nothing lost, and put reporting in front of people who open it.

Pipeline health, one working day

Readings ingestedMQTT, CoAP and HTTP
Rows reconciledSource and destination counts match
Query on a year of historyTime series storage
Failed batchRetried, then escalated
Report refreshedBefore the morning meeting

Boring when it works, which is the point.

ZeroRecords lost in the Norwegian migration
YearsOf readings kept queryable on IoT Workz
3 weeksOne feed ingested, stored, monitored and reported end to end
3 to 4 monthsA platform across several sources, with migration behind it
What we build

From source to report, with the pipeline watching itself

A data platform, in the order it usually matters
Sources
DevicesMQTT, CoAP and HTTP
Systems and filesAPIs and scheduled drops
Move
Kafka logo
KafkaWhere volume or decoupling demands it
Scheduled batchWhere streaming is not needed
Store
TimescaleDBReadings that arrive continuously
PostgreSQLRelational work
Object storageRaw payloads
Use
Power BI and GrafanaWhatever your team already opens
Pipeline monitoringPrometheus alerts when a feed stops

Migration runs as its own workstream, with row counts and checksums on both sides.

Fit check

Is the problem in the plumbing?

Pick us when

  • Two departments quote different figures for the same thing
  • Data arrives faster than anyone can process by hand
  • You are moving off a system past end of life
  • An analytics or AI project stalled waiting for clean inputs

Look elsewhere when

  • A spreadsheet is still fine
  • One person exports one report once a week
  • Nobody disputes which number is correct
  • You need depth on Snowflake, Databricks or BigQuery, which we have not shipped on
How the work runs

One feed end to end before we widen

From source map to handover
Week 1

Map the sources

Where data comes from, how much, how often, and which number people argue about.

Weeks 2 to 3

One feed, end to end

A single source ingested, stored, monitored and reported, so the shape is proven first.

Weeks 4 to 12

Widen and migrate

More sources, then any migration, rehearsed before the real run.

Handover

You run it

Pipelines, monitoring and documentation are yours, deployed in your own tenant.

Reconciliation and monitoring are the two steps people skip, and they matter more than ETL against ELT.

Technologies

What we build on

Everything here runs somewhere in production. We left off tools we have only read about.

PostgreSQLKafka logoKafkaPower BI
TimescaleDB
MQTT
MinIO
Grafana
Prometheus
Builds

Two jobs where the data was the project

IoT Core Engine self hosted platform built by Dev Station
Case study, IoT platform

Readings that never stop arriving, stored so they stay useful

We built the ingestion path over MQTT, CoAP and HTTP and put readings into TimescaleDB, with Kafka where decoupling mattered and Prometheus and Grafana so the pipeline reports on itself.

YearsHistory kept queryable
National education analytics platform rebuilt by Dev Station
Case study, Norway

Moving years of school records while schools kept working

The client had to leave a system that could not meet its volumes or security standards. Migration ran as its own workstream with a dedicated team, reconciled on both sides, with 60 days of warranty after go-live.

ZeroRecords lost in the migration

Tell us where the numbers break

Three fields. An engineer reads it, and you get an answer within one working day.

  • We map sources, volumes and who disagrees with whom
  • You get a written view on what to fix first, with a cost band
  • No obligation, and no phone number needed to start
FAQs

Questions we get on the first call

Do we need a warehouse or a lakehouse?

Usually neither at first. Most of the value comes from reliable ingestion and one agreed source for the numbers people argue about.

Can you migrate us off a system nobody supports any more?

Yes. We treat migration as its own workstream with reconciliation on both sides, and we rehearse it before the real run.

How do we know a feed has broken?

The pipeline monitors itself and alerts. Noticing a flat chart in a meeting does not count as monitoring.

Is this the first step towards AI?

Almost always. Models fail on their inputs far more often than on their algorithms, so a stalled AI project usually points here.

Tell us which number people argue about

That argument usually points straight at the broken feed. Thirty minutes and you will know where to start.

Let's Talk