Data in motion

A traditional data platform collects data after the fact and tells you what happened yesterday. We build flows that capture events as they happen. That might be signals from a machine on the factory floor or transactions from a checkout in a store. The data moves through streaming pipelines to alerts, decisions and applications that react within seconds, not the next morning.

We work in two areas. The first is industry and industrial IoT: we connect machines, PLCs and control systems, give the signals structure and context, and link production data to orders, batches and quality outcomes. The second is real time beyond the factory; retail, logistics, finance and other businesses where minutes matter.

In both cases we take responsibility for the whole chain: from architecture, edge and connectivity to stream processing, storage for both real-time and historical data, and the interfaces where people and systems actually use the data.

Where delay costs

Often it's the delay that costs. The data was there, it just arrived too late.

Anomaly detection on critical equipment. Vibration, temperature and power consumption are analysed continuously against each machine's normal behaviour. A bearing that is starting to fail, or a process drifting out of its window, is detected hours or days before the breakdown, while there is still time to plan a stop.

Real-time inventory. In retail and logistics, stock levels update across every channel the moment something sells. The online store doesn't sell items that have already run out, and replenishment is driven by what is actually selling right now.

‍

We build for production, not for a demo. That means monitoring, error handling, data formats that can change without breaking anything else, and a solution that survives the network going down on a Tuesday night.

The real-time flows land in the data platform you already have, whether that's Snowflake, Databricks or Microsoft Fabric. Streaming and historical data become one whole: the same data drives today's alert and next quarter's analysis. [Link: Data Platform]

We prefer open protocols and standards. That way you own your data and your architecture, and you can change vendor later without rebuilding.

Technologies and systems we work with

Real time makes different demands than batch. Events have to be handled in the right order and at the right time, state has to survive restarts, latency is measured in milliseconds, and no data can be lost when a link drops. We choose technology layer by layer based on what the environment requires, in the cloud, on-prem in the factory, or in a fully air-gapped OT network.

Edge and connectivity — where data is born. OPC UA, MQTT with Sparkplug B and Modbus towards machines and control systems, organised in a Unified Namespace so that every signal has a name, a unit and a context. Lightweight, robust edge services with local buffering (store-and-forward), preferably in Rust where low resource use and stability are critical.

Streaming and processing — where data moves. Apache Kafka and Azure Event Hubs as the backbone, with schema registry and clear data contracts between producers and consumers. Apache Flink and Spark Structured Streaming for stateful processing with event time, windowing and exactly-once semantics. Change data capture with Debezium pulls events straight out of business systems such as ERP and MES.

Storage and analytics — where data becomes answers. ClickHouse and other columnar databases for fast queries over large time series. Databricks, Snowflake and Microsoft Fabric Real-Time Intelligence connect real-time data with the rest of the business's data, with open formats such as Delta Lake and Apache Iceberg underneath.

Operations and platform — what keeps it running. Kubernetes in the cloud or on bare metal in the factory, infrastructure as code and GitOps, and observability with Prometheus, Grafana and OpenTelemetry.

The hub. Industrial is led by Niclas Grahm, For more than ten years, Niclas has been building data platforms. In recent years, almost everything has come down to the same thing: data that needs to be used while it's still fresh. Sensor data from a production line. Sales ticking in mid-campaign. Decisions that can't wait for tomorrow's report.
‍

Data that exists in the machine but never leaves it. PLC tags without context, historians that can only be read locally, and production data that can't be linked to the order or batch it belongs to. Nightly batch jobs answering questions where the answer was needed within minutes. Pilots built on a script on a laptop on the production floor, which never got further because nobody knew how to run them in production.

And more and more often: an organisation that wants to use AI for predictive maintenance, process optimisation or forecasting, but where the data underneath doesn't hold up. A model on time-series data is never better than the data it is trained and run on. The data has to be correctly timestamped, have context, and arrive in time. [Link: AI & Machine Learning]

That's where we make the biggest difference, when data moves as fast as the business.

Let's talk data.We'll help you find the right way forward, wherever you are today. Contact Niclas Grahm.

‍