Published:
If you are running standard web applications on Kubernetes, the environment feels like a high-security facility. If you allocate 1GB of RAM to a pod, the Linux kernel acts as a relentless enforcer; the moment that pod attempts to touch 1.1GB, it is instantly terminated (OOMKilled). Similarly, CPU cycles are metered with surgical precision using Completely Fair Scheduling (CFS) quotas.
Published:
In Part 1 we explored how Kubernetes threats GPUs as a monolithic integer resource. For a massive training job, this may be a good fit. For a lightweight inference server utilizing only 2GB of an 80GB A100, it is a staggering waste of capital.
Published:
In the evolving landscape of software engineering, we are witnessing a pivot from “Copilots” to “Autopilots,” or agents that work while you sleep. Leading this charge is a methodology with a name as chaotic as it is brilliant: The Ralph Wiggum Technique.
Published:
In our previous deep dive, we explored the Ralph Wiggum Technique , a method defined by its beautiful simplicity. It was the software equivalent of an infinite monkey theorem: set up a bash loop, feed an error log to an AI, and let it fail its way to success.
Published:
In Part 1 we looked at how Harvester handles GPUs, exploring Passthrough vs. vGPU techniques. In this article we will deep dive into how I setup my AMD Radeon 680M to be visible to my VMs in Harvester.
Published:
Published:
In Part 1, we talked about the shift from chatting with AI to orchestrating it. We introduced gemini-cli and the concept of extensions as shipping containers for AI personas.
Published:
Just like everyone else, I’ve been using LLMs daily for coding. I have my library of “perfect prompts,” my favorite web UI, and I thought I had this AI thing figured out.
Published:
