andrew.dunn.dev

Talks

Talks

A boot chain drawn as a row of links, firmware to bootloader to kernel to initramfs to root filesystem, with the links an image can actually enforce at day 0 gathered inside a bracket and the rest of the chain left open after it.

Melbourne Red Hat User Group

Melbourne, Florida · October 21, 2026

  1. From the homelab to your lab

    A field report from the homelab: Fedora bootc images built and boot-tested in GitLab CI, a sealed root checked on read, and our own keys, so servers update like a phone. What it cost, what broke, what is still open.

    1. DeckSlides, in the browser or as a PDF: phone-style updates for servers.
    2. WalkthroughDraftIn preparation: the talk as a scroll-through read, from containers and bootc to boot gates, TPM trust, and landing keystone.
    October 21, 2026
Cover for the AI in Systems Engineering guest lectures: the course drawn as three motifs side by side, the agent loop as a ring of four stations, the cost ladder as rungs read from the cheapest answer up to the least reversible one, and the parse-to-index pipeline running from plain text into a typed index.

Wayne State University

AI in Systems Engineering, guest lectures · Oct 19 – Nov 2, 2026 · 4 parts

Four talks, one argument: what an agent actually is, what it takes to trust one with engineering work, what structure that work needs underneath it, and how to build the bench that measures which of it pays.

Read them in order. Each talk ends where the next one begins.

  1. Hands and feet

    What an agent is, and why git matters. Talk one of four for the course: the loop, the context window, the folder that loads first, and the two gates.

    1. DeckSlides with speaker notes, in the browser: from language model to agent.
    2. WalkthroughThe talk as a scroll-through read: what an agent is, the folder, the two gates.
    3. Lesson · LarderThe build, chapter by chapter: needs Claude Code and a folder, no coding.
    October 19, 2026
  2. The ladder and the meter

    AI in engineering practice. Talk two of four: meta-models, graphs, retrieval, stuffing, and the evaluation loop that earns each rung.

    1. DeckSlides with speaker notes, in the browser: from ontology to meta-model.
    2. WalkthroughThe talk as a scroll-through read: context, retrieval, tools, and the scorer behind each number.
    3. Lesson · OntologyThe kitchen model, built and script-checked; needs a folder and a coding agent.
    October 26, 2026
  3. Structure over the trove

    Meta-models, SysML v2, and one question asked four ways. Talk three of four, leading into the Representation lesson.

    1. DeckSlides with speaker notes, in the browser: one model handed over four ways.
    2. WalkthroughThe talk as a scroll-through read: meta-model to parsed graph, into the Representation lesson.
    3. Lesson · RepresentationAn interactive page for desktop: recorded runs with no key, or live with yours.
    November 2, 2026
  4. A Bench of Your Own

    Take an LLM step you already run and build the evaluation that tells you which rungs of the ladder pay. Talk four of four: your own bench.

    1. DeckDraftIn preparation: a stub deck, exhibits planned, for the bench and the ladder.
    2. WalkthroughDraftIn preparation: a stub of the read, half evaluation design, half the ladder.
    November 2026