The Reanimation
Process
No one is genuinely being raised from the dead. V1 is a retrieval-first system: download public-domain texts, embed them, retrieve relevant passages, and ask the model to answer in a constrained historical voice.
Exhume the works
Each figure begins with public-domain, first-person-heavy sources selected for rights clarity, quality, and citation coverage.
Embed the corpus
The ingestion pipeline chunks full source works, creates embeddings, and stores prepared passages with citation metadata for retrieval.
Retrieve, then speak
Before a reply is written, the agent searches the persona library and returns citeable source links with the answer.
Shape the voice
Prompt and style guides carry V1. Fine-tuning stays optional until evaluations prove it improves voice without weakening citation fidelity.
Discovery to publish
What each mind is grounded in
0 full source works and 0 citeable passages are loaded for the active V1 figures.
Things we will cheerfully admit
- These are AI impressions, not resurrections.
- Accuracy improves as each source corpus gets broader and cleaner.
- If the retrieved sources do not support an answer, the agent should say so instead of inventing confidence.
- Fine-tuning is a later experiment, not a V1 dependency.