AI
Working notes and talk material about building and running a self-hosted AI "software factory" — a fleet of coding agents, the evaluators and gates that keep them honest, and the economics of running the whole thing on your own hardware. Everything here lands ahead of the videos: these pages are the planning notes the episodes get scripted from, posted as-is.
If you only look at one thing, make it the Fleet Explainers — five animated diagrams of how the fleet actually works, failure paths included.
Building the Fleet
- Fleet Series — working notes for the video series: one overview episode, then a deep dive per layer. Start with 01 — The Big Picture; the thesis lives in 05 — Gates: the Evaluator Is the Product.
- Fleet Explainers — the series' visuals: five self-contained animated plates (tasks, one worker's life, waves, the reconcile barrier, the whole machine), each with a full text transcript.
The Bill
- Cost Series — the sibling track on LLM economics: stop counting tokens, price the work. Caching, routing, output discipline, and cost-per-success.
Talks & lens pieces
- Talks — the raw material the series grew out of: a 5-minute lightning talk with full scaffolding, a short opinionated take on agentic-loop hype, and speculative notes reading the fleet through Minsky's Society of Mind.
From the articles & blog
2024-06-20— Resume Bot
I decided to test out a little resume bot using Langflow, a visual tool for building RAG applications.