AI Ingestion & Synthesis Engine

Multi-Model Strategy

1. Cloud LLMs (OpenAI GPT-4o / Claude 3.5 Sonnet):
   • Primary synthesis of multi-source historical archives, council submissions, and community reviews into evocative narrative scripts.
2. On-Device SLMs (Google Gemma 2):
   • Zero-connectivity contextual summaries executed locally on client hardware.
3. Pluggable Audio Dispatcher:
   • ElevenLabs: High-fidelity, emotional narration for sacred cultural stories and curated historical highlights.
   • Device Native TTS: Zero-data, low-latency playback for commercial reviews and offline routes.

Automated Scraping Workflow

A weekly serverless pipeline executes the following loop:

1. Queries OpenStreetMap Overpass API for amenities (bakery, pub, lookout, historic) within targeted coordinate bounds.
2. Cross-references top community ratings and historical notes.
3. Synthesizes 30–45 second driving-friendly scripts.
4. Inserts draft records into yonda-cms marked with source: ai_generated awaiting curator approval.