Blog

I almost agree with all of the LLM critics, yet I still use LLMs a lot. I know this sounds like I am delusional, but I don't think I am alone with it.

Discussion on HN — 310 comments

How CFS throttling works and the case against the average CPU graph.

One in twelve packs of smoked salmon in German supermarkets has detectable listeria. And it's not from the fish.

tokens go brrrr

Why UMH exists, how we work, and what we build.

Beyond compressed air and electricity, data should be the third resource every factory runs on. Today, it's blocked. We raised €5M to change that.

A guide for LLMs (and humans) to write in my voice. Self-demonstrating.

How to properly use LLMs to prevent producing slop.

A Christmas Eve journey. Deutsche Bahn. 35 kilometers.

Discussion on HN — 1008 comments · LinkedIn — 20 comments

Master Industrial IoT downsampling in minutes: compare lossless compression to lossy tricks like report-by-exception, dead-band filtering and swinging-door trending. Learn how UMH-Core applies them at the edge to slash Unified Namespace bandwidth and historian storage by up to 98 %.

Why most factory-floor LLM demos break apart in production, and how a generate → freeze → test workflow restores trust.

Discussion on LinkedIn — 54 comments

Frustrated by confusing MQTT explanations? This guide clarifies MQTT’s essentials, shows why it’s ideal for IoT telemetry, and reveals its drawbacks for manufacturing—plus how to integrate MQTT with data-center-grade solutions for end-to-end reliability.

Discussion on LinkedIn — 40 comments

Writing technical articles is an art. This is wrong. It's a methodological process.

Why we recommend Hetzner to try out the UMH: it is at least 4x–10x less expensive than AWS and Azure for similar VM instances.

Discussion on HN — 138 comments

We're building an integrated and user-centric solution for industrial IoT. This article explores the shortcomings of typical manufacturing software and how UMH's approach addresses them.

Read this article to know how modern "cloud-native" technologies such as Kubernetes, Grafana, and TimescaleDB can and must be combined with the traditional automation pyramid.

Co-authored with Denis Gontcharov. Why we chose Grafana over traditional industrial visualization tools.

Three reasons why OPC UA is not delivering on its promises: it is unreliable in practice, a security nightmare, and standardizes without actually standardizing anything.

Discussion on LinkedIn — 73 comments

A four-chapter walkthrough of the Unified Namespace: how it slots into existing OT systems, where the term came from, how it grounds in established IT principles, and concrete examples from the field.

Discussion on LinkedIn — 41 comments

The article looks at the different data requirements of frontline workers and business analysts, the roles of OLTP and OLAP databases, and the concept of Lambda Architecture for managing large volumes of real-time and historical data.

Explore the pros and cons of Historians versus open-source databases in sectors like oil&gas, pharmaceutical, and chemicals. This article discusses compression algorithms, storage costs, and modern compatibility.

Discussion on LinkedIn — 49 comments

Navigating the design of IT/OT infrastructure without MQTT, Kafka, and Kubernetes? We embarked on that journey too and confronted numerous challenges. Our hope is that our experiences can provide you with insights towards better solutions.

Discover why we chose Flatcar as the operating system for the Industrial IoT. We want to guide you through our thought process - from our requirements to selecting the operating system.

Are you looking for the best MQTT broker for your IoT or IIoT project? In this technical comparison, we evaluate four popular options: Eclipse Mosquitto, VerneMQ, EMQx, and HiveMQ. We consider factors such as reliability, scalability, and maintainability to help you make an informed decision.

Discussion on LinkedIn — 29 comments

OSIsoft PI, Canary, InfluxDB, TimescaleDB and Ignition - the number of tools you can use to store your data from your machinery is endless. This article provides an overview of historians and open source time series databases.

The industry is shifting from use-cases to being technology driven. We provide a look behind the United Manufacturing Hub and into our experiences for tools & techniques for data processing.

How an open-source tool is establishing itself in a highly competitive environment against billion-dollar companies.

Discussion on HN — 28 comments

TimescaleDB is better suited for the Industrial IoT than InfluxDB, because it is stable, mature and failure resistant, it uses the very common SQL as a query language and you need a relational database for manufacturing anyway