arrow_back Back to Dashboard

AI Prompts

This page displays the exact prompts used by FashFinder's AI classification system. These prompts are automatically extracted from the Python source code, ensuring what you see here is exactly what the AI sees when analyzing content.

sync This page is auto-generated from backend/classifier.py and backend/patterns/llm_analyzer.py

Overview

FashFinder uses a multi-stage AI pipeline powered by DSPy (Declarative Self-improving Python). Each stage has specific instructions that guide the AI's analysis:

Stage 1: Initial screening (filters ~80% of benign content)
Stage 2: Category identification (identifies 1-3 matching categories)
Stage 3: Deep classification with quote extraction and severity rating
Stage 4: Cross-category pattern analysis
Stage 5: Event grouping (clusters related incidents)
Weekly: LLM narrative analysis with historical context

Classification Pipeline

Pattern Analysis

After individual classifications are generated, FashFinder runs a separate AI analysis to identify systemic patterns, assess escalation risks, and provide historical context.

Why are these prompts public? Transparency is critical for accountability. By exposing our exact prompts, researchers and journalists can evaluate our methodology, reproduce our work, and understand exactly how the AI arrives at its classifications.