aloth/RogueGPT

Implements a three-layer architecture (core data layer, Streamlit/CLI/MCP interfaces) for generating diverse multilingual news fragments across 37 LLM configurations from 10 providers, with declarative prompt templates and MongoDB persistence. Integrates with the Model Context Protocol for AI agent tooling and connects downstream to the JudgeGPT human perception evaluation pipeline. Dataset includes 2,663 AI-generated fragments plus 222 human-sourced anchors sourced from RSS feeds and credibility-scored fake news domains, enabling controlled experiments on LLM-generated misinformation detectability across languages, styles, and formats.