Lumina
Free preview · read before you buy

Read three lessons free. Then get the full harness.

No signup. Three complete lessons — the vocabulary, the method's first move, and the method applied to security. Real prose, real diagrams, real code. If they're worth your time, the full line is.

The method: generate → judge → gate → retry. LLM is the commodity; the harness is the moat.

Start here

~30 minutes end-to-end. If you can read Python, you're ready.

LEVEL 0 · PRIMER

Before you start

The four ideas every module assumes — an LLM call & a trace, RAG, the Python you need, and judge / gate / eval. No eval or ML background required; the jargon is taught from zero.

Read the primer →
0.1 · THE METHOD

Start from failures — triaged by a judge

The first move of the whole method: point a discovery judge at your traces, get back a ranked failure taxonomy, and decide which failures deserve a gate. Read the real code here; the runnable lab ships with the course download.

Read lesson 0.1 →
SEC.1 · THE METHOD, APPLIED

The threat map

The same method pointed at security: map an LLM app's attack surface to the OWASP-LLM Top 10 and spot the blind spots — the first move of the Securing LLM Apps track.

Read lesson SEC.1 →

Like the method?

The full line is 8 tracks · 58 lessons · 57 runnable, offline labs · a research brief per track — a downloadable kit you own, not a stream. Explore the syllabus while launch details are finalized.

See the full syllabus →