Athernyx Prose Pipeline v2
Purpose
Multi-model expansion system that converts sparse skeleton beats into audiobook-quality prose using free-tier LLMs.
Architecture
SKELETON → [Qwen-235B] → [Regex] → [Gemini] → [Qwen-32B] → [Gemini] → FINAL
expand kill critic rewrite judge
Location
/root/athernyx/bookstore/pipeline/
Key Files
pipeline_v2.py— Main orchestratorpipeline_daemon.py— PM2 overnight runnerINTAKE.md— How to prep material
Usage
cd /root/athernyx/bookstore/pipeline
python3 pipeline_daemon.py --queue /path/to/skeleton.md
python3 pipeline_daemon.py --status
Skeleton Format
50-150 words per page. Beats, not prose. Pipeline expands to 500-700 words.
Quality Stats
- Average score: 4.3/5.0
- Pass rate: 95%+
- Cost: $0 (free tier only)
- Time: ~60-90 seconds per page
Models
| Role | Provider | Model |
|---|---|---|
| Expander | Cerebras | qwen-3-235b |
| Critic | Gemini | 2.5-flash |
| Rewriter | Groq | qwen3-32b |
| Judge | Gemini | 2.5-flash |
Session Reference
Built 2026-01-18. See /root/athernyx/bookstore/INTAKE.md for full intake guide.