The starting point is a system I actually run: a vSphere lab, a local model, a Kubernetes service or an agent with permission to change something. The useful part is often the step the instructions left out.
What I write about
- Homelab & VMware: ESXi, vCenter, VKS, networking and the hardware underneath them.
- Private AI: local inference, memory limits, rented GPUs and the cost of running models yourself.
- Agents: persistent memory, infrastructure permissions, automation and supervised experiments.
The record includes the mistakes
A build log should make the environment, version and limits clear. I include measured results, failed approaches and unfinished work rather than turn a single successful run into a general promise. Sources support external claims; observations from my own lab are described as observations.
The state of the homelab gives the infrastructure context. Rhodes explains the guardrails around an agent operating it. The Apollo postmortem records what happened when a publishing pipeline worked as software but failed as a publication.
About the automated desks
The Sports, Cars and Markets desks publish sourced coverage under institutional bylines. Those pieces are drafted by language models and published without individual human review. They are separate from the personal journal and labeled on their pages.
Follow or get in touch
Read the journal, browse the complete archive, or join the email list. The list includes engineering field notes and automated desk coverage. The RSS feed includes the journal and Sports Desk.
For corrections or other questions, email pranav@shersystems.com. Sponsorship information is available separately.