DevocionalesAPI

This repository generates biblical devotionals using large language models. It contains server components for single-item generation and a provider-agnostic batch pipeline for yearly/devotional generation from seed files.

Quick links

• GEP (Genome Evolution Protocol): GEP_Genome-Evolution-Protocol/README.md — Quality assurance system with two-phase validation
• Batch pipeline documentation: seed_generation/README_BATCH_PIPELINE.md
• Batch scripts and adapters: seed_generation/
• Server and client: API_Server.py, API_Client.py

Overview

• API_Server.py: FastAPI server that calls an LLM adapter to generate devotionals.
• API_Client.py: Simple client for iterative generation via the server.
• seed_generation/: Batch pipeline to submit seed JSON files to provider adapters, collect results, repair failures, and validate outputs.
• GEP_Genome-Evolution-Protocol/: Quality assurance critic system using simulated readers with evolving pattern genome.

Batch highlights

• Provider-agnostic submission via batch_submit.py and provider_adapter.py.
• --dry-run: write provider-agnostic prompt JSONL for review (no API calls).
• --dry-run-full: print the exact wire payload for one seed entry (no API calls) to inspect adapter formatting.

GEP — Quality Assurance

The Genome Evolution Protocol (GEP) validates devotional quality through a two-phase process:

1. Phase 1 (Linguistic): Fast scan for typos, grammar, repeated phrases, and unnatural phrasing

   • No genome injection for clean, unbiased signals
   • Verbatim gate filters hallucinated flags
   • Returns: CLEAN or FLAG

2. Phase 2 (Content): Deep review for prayer drift, register issues, and hallucinations

   • Genome-seeded prompts with confirmed patterns
   • Simulated reader persona (language/culture-specific)
   • Returns: OK or PAUSE

Recent improvements (2026-04):

• Removed genome injection from Phase 1 to prevent model confabulation
• Added verbatim gate to filter hallucinated flags before audit log
• Fixed parser for Fireworks responses starting mid-<think> tag
• Added corpus_scan() for deterministic Python-based pattern search across full corpus

See GEP_Genome-Evolution-Protocol/README.md for full documentation.

Getting started

1. Install dependencies:

pip install -r requirements.txt

2. Read the batch pipeline docs:

less seed_generation/README_BATCH_PIPELINE.md

If you want, I can convert other README files to English or add more examples.