gtfoo

Live · One person's programme

Indie Degree

A degree I'm building for myself, out of material that was already free — and the machinery that makes the claim checkable.

203 verified sources15 coursesEvidence tiers

The idea

Everything a degree teaches is already free

I wanted to get properly good at AI engineering, and the material for that is already sitting there for nothing — largely the same lectures, papers and textbooks a degree would assign anyway. What a degree sells on top of it is structure and attestation: the right things in the right order, and an institution willing to say you did it. So I built the structure myself.

Two things break the moment you do that alone, and they are the same thing: a claim nobody can verify. Ask a model for a reading list and you get plausible titles and links that resolve to nothing. Award yourself a credential and it carries no weight. So the whole thing runs on two rules — the model may select sources but never invent them, and every claim of mastery has to carry the evidence that earned it.

v1

Verify first, build second

The corpus came before the app, which is the right way round: a tracker for a curriculum nobody has checked is a tidy way to record fiction. So the verifier was built first, and the app that tracks progress came afterwards — the smaller half.

Along the way

What it grew into

  • Identity checks, not link checks

    Every resource declares how its identity can be proven, and the checker proves that rather than settling for an HTTP 200 — a renamed page still resolves, and YouTube serves 200 for deleted videos. A talk is confirmed through oEmbed against its real title and channel, a paper through the arXiv API, a book by ISBN, a repo through the GitHub API. It caught mistakes in hand-written entries too, not just generated ones.

  • Evidence tiers, and tier 0 is worth nothing

    Mastery claims are graded by how they were evidenced: self-marked, machine-verified, panel-assessed against a rubric registered before submission, an artifact in the world, or defended aloud unscripted. Self-marking is tracked and deliberately counts for zero — so “I read the chapter” can be recorded honestly without inflating anything.

  • A programme, not a reading list

    15 courses and 75 skills arranged as a dependency graph, with an advanced-standing file that grants credit for work already shipped.

Trade-offs

The judgment calls I'm proud of

The model selects sources; it never invents them

This is the decision the whole thing rests on. A generated syllabus is worthless if its references are plausible-looking fiction, and asking a model for “good resources on X” produces exactly that. So the model proposes, and nothing enters the corpus until an external API has confirmed the thing exists and is what it claims to be.

A self-awarded credential has to be auditable

Anyone can declare themselves qualified, which is why self-declared learning carries so little weight. The answer wasn’t to make the claims louder but to attach their evidence: a reader doesn’t have to trust the grade, they can see how it was earned.

Where it is now

One degree, in the open

It is my programme, public and read-only to visitors — the curriculum, the sources and the evidence rules are all there to inspect, which is rather the point of a credential that asks to be audited. Progress moves only when the work actually happens, and publishing something that can only embarrass me is deliberate.

The next phase is to let anyone generate their own. Nothing in the machinery is specific to AI engineering — swap the resources and it produces an indie degree in something else. That is the direction, not a shipped feature.

Read the programme

15 courses, 75 skills, and 203 sources you can check for yourself.

Launch the app ↗