Step 1
Load study material
For this first Herman version, the cleanest inputs are `.md`, `.txt`, `.html`, `.csv`, or `.json`. If you have a PDF, paste the useful text below for now.
Core loop
What Herman is doing
Turn uploaded documentation into a local source pack with normalized concepts and evidence snippets.
Generate multiple-choice checks from those concepts to find weak or unfamiliar ideas quickly.
Traverse the weakest branch depth-first, keep misses local to that branch, and weight low-confidence answers almost like misses.
Group source material into topic clusters so the engine can stay inside a weak neighborhood instead of sampling unrelated questions.
The implementation is intentionally static-first right now so this can live comfortably in Herman while the future upload pipeline gets smarter.
Step 2
Run the study engine
Step 3
Gap map
Weakest concepts
Next step
Cluster coverage
Recent sessions
Concept map
Search the current knowledge base
Library