What Should I Show in a Demo?

Use this as a short demo path through the published brain.

Suggested Flow

1. Start at codebase-overview to establish that this is a reproduction repo, not the original exploratory workspace.
2. Open prompt-system to show the source-specific extraction design.
3. Open synthetic-data-generation to show how the repo creates additional structured supervision.
4. Open dataset-and-config-map to show where data and training configs live.
5. Open evaluation-and-decoding to connect the repository to the competition metric.
6. Open missing-283-transliterations if the audience wants provenance from the original working repo.
7. Close with brain-publishing to show that the wiki itself is generated from brain/ and published by Quartz/GitHub Pages.

Demo Framing

The strongest framing is that the solution is reproducible because the repo separates:

• source extraction;
• data normalization;
• synthetic data generation;
• final dataset assembly;
• model training;
• evaluation and decoding;
• provenance notes.

This makes the brain useful as a guided map rather than a replacement for the code.