If I figured it out, I wrote it down.
David Duncan from AWS showed how to run AI agents on Linux using service accounts and Btrfs subvolumes. The architecture solves the memory persistence problem that containers cannot.
Kyle Quest showed a live demo that turned 9,000 container CVEs into zero without fixing a single one. Here's what that means for how we read scanner output.
Brian Ragazzi walked through KwaaiNet, a decentralized AI inference network built on Petals and DHT. The architecture distributes transformer layers across volunteer nodes the way BitTorrent distributes file chunks.
David von Thenen walked through building a nanoGPT-style model end-to-end, covering tokenization, the 10TB streaming problem, fine-tuning, and quantization in a homelab-friendly way.
Notes from the IBM Granite team's Docling workshop at SCaLE 23x: what the library actually does, how chunking works for RAG, and why running it locally matters.
How Tailscale's WireGuard-based zero-trust overlay network replaced a tangle of port forwarding rules and static routes with something that just works, and what that taught me about what networking abstractions actually cost.
How a persistent question about what's actually happening underneath the systems I work with every day turned into a full infrastructure lab running AI agents, containers, Kubernetes, and a security research environment at home.
My job title says support analyst. That's accurate and also incomplete. This is what actually drives me, and why I'm writing about it.