# Daily Audiobook Research

Full production pipeline that turns knowledge base connections into 20-minute audiobooks. Scans for non-obvious cross-domain intersections, picks an unwritten topic, researches deeply, writes 25-35K characters of narrative nonfiction, generates audio, and delivers.

**When to use this**
When you want to produce educational audio content at scale — daily briefings, training material, knowledge transfer. The pipeline is designed for one book per day, fully autonomous.

**The process**

1. **Mine connections.** Load your knowledge base sections and ask: "What does domain A know that domain B desperately needs but does not know it needs?" Each intersection is a potential topic.
2. **Pick a topic.** Priority 1 = directly relevant to active projects. Never repeat a topic already produced. Check the learning log.
3. **Research deeply.** Find 5-10 real examples, studies, or historical cases. Search for what the audience does NOT already know. Find contrarian angles and failure cases. Mine YouTube transcripts for expert lectures and interviews.
4. **Write.** 15+ chapters, 1,000+ words per chapter. Narrative book style, not a lecture. Third person only. No hashtags, no markdown, no em dashes. Each chapter ends with a non-obvious insight that compounds across the book.
5. **Review.** 14 rounds: 7 for repetition (structural, sentence-level, concept rehash, phrasing echo, data redundancy, density, fresh eyes) and 7 for craft (word pictures, narrative arc, example specificity, contrast, transitions, voice, emotional architecture).
6. **Generate audio.** TTS with consistent voice, chunked at natural boundaries. Output as a single file.
7. **Deliver.** One complete file with a companion text message containing title, chapter count, source count, and a hook.

**Pre-prepared sub-agents**

**1. Topic miner** — loads knowledge base sections, finds cross-domain intersections, picks unwritten topics
**2. Research agent** — deep research with 50+ sources, YouTube transcript mining, contrarian search
**3. Writer agent** — narrative nonfiction, 15+ chapters, escalating insight architecture, 14-round review
**4. Audio agent** — TTS generation, chunking, concatenation, delivery

**Tips**
- Each chapter must end with an insight that is UNIQUE to that chapter and compounds on all previous insights. If two chapters could swap places, one is circling, not stacking.
- Every mechanism described must include its outcome: not just "how it works" but "what result it produces and why."
- Privacy rule: never reference specific people, projects, or internal databases in the audiobook text.

**Tags:** Education, Content Creation, Writing