Build Journal: The Story Behind the Stack
Not tutorials — stories. The wow moments, the crashes, and the 2am realisations that come from actually building with AI instead of just reading about it.
Why this journal exists
I started this journey with a box, a graphics card, and a quiet conviction that the best way to understand AI is to live with it. Not in a lab, not in a tutorial, but on a desk strewn with cables, half-drunk coffee, and the occasional frustrated sigh. Build Journal is where I share that life — the unfiltered, human side of building with AI on an RTX 5070 Ti running Ubuntu.
This isn’t a tutorial series. You won’t find step-by-step guides to installing any of the tools or optimizing inference speeds here. What you’ll find are stories. The kind you swap over a fence or a forum thread at midnight: the moments when the model did something uncannily brilliant, the times when the whole stack came crashing down in a spectacular loop, and the quiet lessons learned from mistakes that cost me more than just time.
What I write about
Think of it as my build log, but with heart. I’ll talk about wow moments — like the time where Dade, my Hermes Agent deduced where it got it's name from from a simple link to IMDB without a single prompt. I’ll tell crash narratives, such as the Ollama crash loop that turned my careful experimentation into a frantic dance of restarts and log diving. I’ll share lessons from failure, like the infamous 12 Million Token Mistake that taught me why context window management isn’t just a theoretical concern. And i’ll pull back the curtain on building in public, showing the wiring, the whiteboard sketches, and the honest trade-offs I have to make make as an indie builders on a budget.
Some builds are technical. Some are weirdly personal. Reviving Kate sits somewhere in the middle. She is my “OpenClaw in a box” experiment: a real agent environment with proper guardrails, restricted access, and a clear goal — prove that OpenClaw can help inside a business-style setup without being able to wander off and cause damage. That is why I tightened security, limited what she can touch, and treated the revival as an install-readiness test rather than just another bot coming back online.
Each post is meant to be read with your morning coffee — or maybe your evening tea, depending on when the inspiration strikes. I want you to feel the excitement when a prompt returns something unexpectedly insightful, the frustration when a version change undoes hours of work, and the quiet satisfaction of solving a problem that only appeared because I dared to run the thing by myself.
What is coming
Here’s a taste of what’s coming:
- ‘The IMDB Deduction’: A wow moment where my AI connected dots across obscure film trivia without being asked to do so.
- ‘When Your AI Stack Eats Itself’: A crash narrative about an Ollama-induced feedback loop that brought my local pc to its knees.
- ‘Day 1: The HDD Arrives’: The origin story — unboxing the emtpty Harddrive, the first whiff of new hardware, and the mix of hope and terror as I got it running for the first time.
- ‘Building Mission Control’: How I assembled my monitoring and management dashboard because flying blind is no way to run an AI lab.
I’ve learned that every crash taught me something — and most importantly that every prompt has taught Dade something as well. And every surprise, every ‘wait, how did it know that?’ moment, reminded me why I started: to see what happens when you put powerful tools in the hands of curious builders and let them explore. This journal is my invitation to join me on the journey, not just as spectators, but as fellow tinkerers.
Welcome to the workbench
So welcome to Build Journal. Pull up a chair, ignore the stack of manuals in the corner, and let’s see what happens next.
Found this useful? → Follow @Raf_VRS for more Build Journal updates → Support the work: ko-fi.com/rafvrs #HardInterference #BuildInPublic #AIAgents