Context Jamming · Production Colophon
Architecture of an AI-Native Book
Local CPU inference. No API key. No recurring cost. Model weights download once (~330MB).
British male, measured editorial register. Closest open-source analogue to New Yorker audio narration.
Phonemizer backend, WAV→MP3, chapter stitching with 3.5s silence, NumPy I/O at 24kHz native.
AI-hosted audio explainer embedded on the TOC. Two AI hosts walk the full arc of the book.
public/book/ and public/audio/ — ~190MB of HTML and MP3 assets committed to git and served from CDN edge."Copied!" feedback, copper accent paletteCHAPTER_VIDEOS + CHAPTER_PRESS maps — per-chapter media registry in build.ts; adding new links is a one-liner<filename> tags for inline markup