Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.pcm.game/llms.txt

Use this file to discover all available pages before exploring further.

​
The AI Content Factory

The system relies on an asynchronous background pipeline powered by Evergreen RSS Feeds to ensure the database is constantly hydrated[cite: 591, 592].

​
1. The RSS Ingestion Worker

Cloud Scheduler triggers a Python FastAPI Cloud Run service every 4 hours[cite: 626, 627, 464]. It polls high-reputation RSS feeds across various categories (Hard News, Business, Culture, Tech) and compares incoming headlines against existing stubs using a Jaccard similarity algorithm (dropping anything ≥85%\ge 85\% similar)[cite: 602, 619, 620].

​
2. Gemini 3.1 Flash-Lite Mapping

Deduplicated stubs are routed strictly to Gemini 3.1 Lite for classification[cite: 621, 632].
Strict Compliance: To guarantee JSON structures without conversational hallucinations, the pipeline uses Native Structured Outputs by passing response_schema into the API[cite: 1225].

​
3. The Local Fallback

If Vertex AI throws a Safety Filter exception due to satirical content, the system reroutes via SSH tunnel to a local open-weight model (like Llama 3 8B via Ollama)[cite: 493, 495].