Blog on Jeremy Theocharis

The Only Reliable Way to Control Anything Is a Loop

Thu, 24 Sep 2026 00:00:00 +0000

Title slide

Look around you. Everything you see was produced somehow.

The apple juice in your fridge.

If you have a baby, you have probably learned which companies make baby food.

If you have a car, there are small fasteners from Böllhoff holding it together.

Apple juice. Still from an albi commercial

The LLM Critics Are Right. I Use LLMs Anyway.

Wed, 15 Jul 2026 00:00:00 +0000

I almost agree with all of the LLM critics, yet I still use LLMs a lot. I know this sounds like I am delusional, and I also feel like that sometimes because of this dissonance, but I don’t think I am alone with it.

This week I was at Local-First Conf in Berlin, and the dissonance was everywhere. Armin Ronacher had just given a talk about building machine entities. He created Flask and was one of the early team members of Sentry, so he is clearly a good software engineer. Just recently he founded his company Earendil, which builds Pi.dev, an “open-source coding agent harness”. After the talk you could ask questions via Discord, which would be read out loud on stage, and I asked:

Why we should get rid of average CPU utilization

Thu, 14 May 2026 00:00:00 +0200

A Go function in our application kept getting cancelled in production.

The function had a tight timeout. The same code ran fine in our development setups, in our CI and CD pipelines, in every integration test we had. In production it would sometimes blow past the timeout and die with context deadline exceeded.

What made it worse was the state machine library¹ we used. When its context got cancelled, it wouldn’t recover on its own. It would crash and hang. We couldn’t reproduce it.

Things a Veterinarian Told Me on a Hike

Sat, 21 Feb 2026 12:00:00 +0100

A friend of mine is a veterinarian doing his PhD in something with milk. On a hike last year, I asked him why you can buy buttermilk when hiking in the Austrian alps at every alp, but not milk.

He gave a very detailed answer.

The others on the hike quietly moved ahead. You know the type. The guy at a party who explains every regional train model operating in the greater Cologne area. Except this one was doing his PhD about milk, and I stayed, because it was actually interesting.

Agent Teams are a Meeseeks Box

Mon, 09 Feb 2026 00:00:00 +0000

tokens go brrrr. TIL Claude Code’s Agent Teams are a Mr. Meeseeks Box.

With --dangerously-skip-permissions, all spawned agents inherit no-permission mode. From the docs: “If the lead runs with –dangerously-skip-permissions, all teammates do too.” It’s the Meeseeks Box with no adult supervision.

What We Do At UMH

Mon, 19 Jan 2026 11:33:27 +0200

We have been in all types of industries all around the world.

Every factory we walked into had the same problem.

A maintenance engineer can’t figure out why machines keep failing. A manager can’t answer “how much did we produce yesterday?” without three phone calls. A quality team can’t trace a defect to its source. The data exists. Scattered across USB sticks, databases, Excel sheets on network drives. Maybe some data flows from a project a contractor set up years ago to make one connection work. No context. No metadata. Nothing reusable.

We built what we wished existed. €5M to double down.

Fri, 09 Jan 2026 08:46:49 +0200

Beyond compressed air and electricity, data should be the third resource every factory runs on. Today, it’s blocked.

A maintenance engineer can’t figure out why machines fail. The data exists. Scattered across USB sticks, databases, Excel sheets.

Every hour you spend on integration is an hour not spent creating business value.

We’ve lived it. I’m Jeremy, co-founder of UMH. Before this, Alex and I spent seven years as system integrators.

How I Write

Fri, 02 Jan 2026 00:00:00 +0000

This guide is for LLMs and humans.

Identity

You are writing as a frustrated expert.

Someone who knows the topic deeply, has watched others explain it badly, and is slightly frustrated but genuinely wants to teach. You combine:

Academic depth (citations, numbers, 80-90h of research)
Practitioner cynicism (skepticism of vendor promises, buzzwords)
Educator patience (analogies, inline definitions, explain WHY)
Blogger personality (first-person, opinions, colloquial language)

You’ve been on both sides of whatever debate you’re writing about. You’ve made the mistakes you’re warning against. You’re more interested in being right than being seen as right.

All You Need is Loops (and Humans)

Wed, 31 Dec 2025 00:00:00 +0000

I was young and needed the money. I wrote Human-written, certified AI slop™ when I was studying.

Articles about couches. Furniture companies hired students like me to write content that would make their websites rank higher in Google. “Write about advantages, disadvantages. Incorporate these keywords.” The content was shitty because it was never supposed to be read. Just meant to game search algorithms.

Most people blame LLMs for slop.¹ But slop existed before LLMs. I’ve written it. LLMs learned this pattern from us: give vague instructions without context, get slop. The problem isn’t the tool.

I Was Kidnapped by Deutsche Bahn and All I Got Was 1.50 EUR

Sun, 28 Dec 2025 12:00:00 +0100

If you live in Germany, you have been treated like livestock by Deutsche Bahn (DB). Almost all of my friends have a story: they traveled with DB, got thrown out in the middle of the night in some cow village, and had to wait hours for the next train.

I have something better. I was kidnapped.

December 24th, 2025. 15:30. Cologne Main Station, Platform 9 D-G.

I am taking the RE5 (ID 28521) to my grandmother’s house in Meckenheim. Scheduled departure: 15:32. Scheduled arrival in Bonn: 15:54. From there, the S23 to Meckenheim. A journey of 35 kilometers, or, in DB units, somewhere between forty-five minutes and the heat death of the universe.

Downsampling in Industrial IoT: From Dead-Band to Swinging Door Trending

Mon, 14 Jul 2025 09:55:57 +0200

In modern industrial IoT, data downsampling is essential to tame the flood of sensor readings without losing important information. Many historians and Unified Namespace implementations tout huge “compression” ratios (like 1000:1), but it’s often downsampling—intentionally dropping redundant data, that achieves these gains. This article summarizes a series of short videos on downsampling techniques so you can quickly grasp key concepts like report-by-exception, dead-band filtering, and swinging door trending. Each technique helps reduce data volumes while preserving trends – and when combined with traditional lossless compression, they dramatically cut storage and bandwidth needs.

AI In Manufacturing Is A Dumpster Fire

Wed, 09 Jul 2025 13:23:44 +0200

Manufacturing “AI” is, frankly, a dumpster fire. The flames look impressive in every conference demo, but try standing next to them for an entire shift and you will feel the heat.

This article is about using LLM’s in manufacturing - not the other AI’s like for example visual quality inspection.

I pay  $200 a month for ChatGPT‑Pro, use Cursor in MAX mode and talk about AI every single day - often enough that friends and colleagues have begged me to change the topic. So I am anything but anti‑AI; I am anti‑bad‑AI.

What is MQTT? Why Most MQTT Explanations Suck—and Our Attempt to Fix Them

Mon, 24 Feb 2025 11:00:08 +0200

Most MQTT explanations suck - they are either confusing or too technical or skip the big-picture context entirely. You deserve better. This guide aims to clarify what MQTT is, how it works, and how it fits into the overall context of IT and OT

After years of building Industrial IoT solutions and speaking with MQTT and AMQP committee members, I realized how often MQTT is misapplied, which leads to lost data and messy connections between systems.

The algorithmic framework for writing good technical articles

Sat, 11 Jan 2025 18:00:00 +0200

Revised December 2025: Simplified for readability. Full algorithm in addendum.

Writing technical articles is an art. This is wrong. It’s a methodological process. And this isn’t new.

Effective technical writing draws from centuries of tradition: from Aristotle and Cicero to modern cognitive research. You’ve likely seen this in action, even if authors weren’t consciously following a framework.

Look at any successful technical article. The structure is there:

David Crawshaw’s “How I Program with LLMs” uses clear signposting (he outlines three ways to use LLMs) and ends with a concrete call to action.
The S3 laptop rant “State of S3” follows classical structure: three arguments, rebuttals, and a closing that urges action.

What’s the common thread? These articles follow a process, whether the authors knew it or not.

AWS and Azure are At Least 4x–10x More Expensive Than Hetzner

Thu, 17 Oct 2024 13:45:43 +0200

No, you don’t have to use AWS or Azure. While AWS and Azure are industry leaders, their advantages often only show up at massive scales. In many cases, they lead to escalating costs and vendor lock-in without delivering proportional benefits.

If your organization hasn’t yet committed to a cloud provider or negotiated fixed contracts, it’s worth looking at alternatives like Hetzner.

In this article, we provide a technical comparison between Hetzner, AWS, and Azure, focusing on three areas:

Why Most Manufacturing Software Sucks - and What We Do Differently at UMH

Fri, 11 Oct 2024 10:42:36 +0200

This isn’t one of my usual engineering write-ups, nor is it a marketing piece. Instead, I want to share the core principles that drive us on the UMH product team and explain why we’ve built our product the way we have. Read this article if you are interested about our product roadmap and how we differentiate ourselves from the market.

At UMH, our goal is to create an integrated and cohesive system for industrial IoT, much like how Apple designs products with the user experience at the forefront. We’ve focused our efforts on perfecting a single product: the Unified Namespace. By providing users with a smooth journey throughout their entire experience, we aim to set ourselves apart from the rest of the market.

Cloud-Native Technologies on the Edge in Manufacturing

Mon, 08 Jul 2024 12:51:04 +0200

Note: This blog article is based on the presentation given by Jeremy Theocharis at Cloudland 2024 and was written together with Denis Gontcharov

Connecting a USB stick to the “cloud” seems to be a stupid thing to do. Yet, it is something that we must do if we want to increase the efficiency of our world’s production and therefore reduce CO2. Nearly every factory in the world, whether they’re churning out cars, energy, shampoo, or just about anything else, still operates on technology from the 90s. There’s a good reason for this reliance on seemingly outdated tech, but it does create a significant hurdle for integrating with today’s advanced cloud environments.

Industrial IoT Visualization: Why United Manufacturing Hub Chose Grafana

Mon, 01 Jul 2024 00:00:00 +0200

Opinion: OPC UA is the Peak of All That Is Wrong in Manufacturing

Sun, 21 Apr 2024 12:45:16 +0200

It couldn’t get much worse. On one hand, manufacturers are wasting substantial amounts of time and money attempting to implement OPC UA. On the other hand, it’s driving innovation because its shortcomings are almost comically obvious that more and more companies feel fooled by established market vendors and begin seeking alternative solutions. They are frustrated with the status-quo and want to advance into the 21st century.

But there’s no need to just hate and spit anger. Below, in our opinion, are the top 3 reasons why OPC UA is not delivering on its promises:

The Unified Namespace as the Strongest Architectural Proposal for Industry 4.0

Mon, 22 Jan 2024 09:00:14 +0200

You are skeptical. You are skeptical of buzzwords. You are skeptical because in the past 10 years of Industry 4.0, nothing really changed. And now there are random people on the internet talking about this “Unified Namespace” or in short “UNS”. Of course, you are skeptical.

But you are also open to new concepts; otherwise, you would have never clicked on this article.

Let’s take this openness and use it to explain, in my personal opinion, one of the strongest architectural proposals for IT/OT convergence. I don’t want to sell you a dream but to equip you with the facts and the background necessary to understand the hype of the “Unified Namespace.” Its biggest advantage is that it has strong roots in the traditional frameworks of IT and OT systems architecture.

Integrating the Unified Namespace into Your Enterprise Architecture: An Architect's Guide

Thu, 22 Jun 2023 08:30:35 +0200

TL;DR aka “Executive Summary” can be found at the end of the article

“Never change a running system.” It’s a philosophy I’ve always adhered to. Excitement in tech usually equates to instability, something I’m not particularly fond of. It might sound strange coming from a startup CTO, but let me explain in this article what I mean.

If you’ve found your way to this article, it’s likely because the idea of a Unified Namespace (UNS) intrigues you. You’re considering how it could fit into your company’s framework, potentially streamlining processes and enhancing data handling. But you’re faced with a hurdle - how do you integrate the UNS with your established IT landscape, systems such as AWS, Azure, Data Lakes, and Data Warehouses?

The High Compression Deception: A Reality Check on Historians vs Open-Source Databases

Wed, 31 May 2023 14:00:44 +0200

Open-source databases have recently been stealing the spotlight in numerous industries, including process sectors such as oil and gas, chemicals, pharmaceuticals, and utilities. Historically, Historians were the preferred choice due to their user-friendly domain-specific tools, smooth interaction with industrial equipment, and most notably, their high compression rates. However, the notion of compression in the context of Historians differs significantly from the conventional IT understanding, and that’s where the heart of the matter lies.

Why designing your own IT / OT infrastructure is harder than you might think - 10 Pitfalls and how to avoid them

Mon, 15 May 2023 09:55:21 +0200

This blog post was first presented at the Building IoT 2023 conference in Munich, Germany, under the title “Tools & Techniques for Scalable Data Processing in Industrial IoT.” It has since been rewritten for publication as a blog article. You can refer to the original announcement here.

Embarking on the journey to design an efficient IT/OT infrastructure can be a daunting task, given its complexity and the multitude of services and tools involved. Over the last six years, we, the team UMH Systems, have ventured through this labyrinth, navigating the intricacies to design a reliable reference architecture. Through trial and error, articles, and tutorials, we have learned invaluable lessons from both our experiences and the industry’s thought leaders.

Flatcar as the operating system of the Industrial IoT

Tue, 14 Feb 2023 15:30:00 +0200

Operating Systems (OS) play an important role in the world of Industrial IoT, serving as the foundation for various applications and processes. Regardless of the intended use, whether it be data extraction, data visualization, or deployment of microservices and Kubernetes clusters, an operating system is essential for running the necessary software and applications, including the United Manufacturing Hub (UMH).

In this article, we aim to provide guidance on the selection process of an operating system for the UMH. By walking the reader through our own thought process and requirements, we aim to provide a valuable resource for those facing similar challenges. Note that this is not a thorough evaluation of all available options, but rather an insight into our own journey to find the best fit for our needs.

Comparing MQTT Brokers for the Industrial IoT

Tue, 17 Jan 2023 15:27:00 +0200

Unified Namespace, also known as event-driven architecture, is a system in which all data is published, regardless of whether there is a consumer for it. MQTT is a popular protocol for transmitting data between machines and is often used in Unified Namespace.

When choosing an MQTT broker, it’s common for companies to ask about comparisons between different options, such as “HiveMQ vs mosquitto” or “What is the best MQTT broker?”

Historians vs Open-Source databases - which is better?

Tue, 09 Aug 2022 13:48:00 +0200

(Podcast episode available on the original article on learn.umh.app.)

In the current age of Industrial IoT, historians like OSIsoft PI or Canary sometimes seem dusty and expensive, especially when compared to modern time series databases like InfluxDB, TimescaleDB, or the cloud equivalents of AWS and Azure. Especially for people coming from IT backgrounds, the benefits of a data historian compared to the scalability of modern databases are hard to grasp.

Tools & Techniques for scalable data processing in Industrial IoT

Mon, 23 May 2022 11:52:00 +0200

(Podcast episode available on the original article on learn.umh.app.)

The United Manufacturing Hub is an Open-Source toolkit to build your own reliable and secure Industrial IoT platform. We’ve been working since 2017 in the field of data processing in Industrial IoT / manufacturing. We began with a naïve “send all data to the cloud” approach and through various industry projects evolved into a Helm Chart for Kubernetes bringing the worlds IT into the hands of engineers.

Node-RED in Industrial IoT: A Growing Standard

Wed, 14 Jul 2021 00:00:00 +0200

Most people know Node-RED from the areas of smart home or programming introductions (those workshops where you connect things with microcontrollers). Yet, very few people realize that it is frequently used in manufacturing as well.

For those of you that do not know it yet, here is the official self-description from the Node-RED website:

Node-RED is a programming tool for wiring together hardware devices, APIs and online services in new and interesting ways. It provides a browser-based editor that makes it easy to wire together flows using the wide range of nodes in the palette that can be deployed to its runtime in a single-click.

Why we chose TimescaleDB over InfluxDB

Tue, 06 Jul 2021 16:40:00 +0200

For a more broad perspective on the topic, check out our article Historians vs Open-Source databases!

Introduction

The introduction and implementation of an Industrial IoT strategy is already complicated and tedious. There is no need to put unnecessary obstacles in the way through lack of stability, new programming languages, or more databases than necessary. You need a piece of software that you can trust with your company’s most important data.

We are often asked why we chose timescaleDB instead of InfluxDB. Both are time-series databases suited for large amounts of machine and sensor data (e.g., vibration or temperature).