LLM-based pipeline for detecting 23 propaganda and persuasion techniques in text. Built on DSPy + OpenAI, with consistency voting, an LLM judge, and span-level visualizations.
