Skip to content

Latest commit

 

History

History
78 lines (63 loc) · 3.03 KB

CLAUDE.md

File metadata and controls

78 lines (63 loc) · 3.03 KB

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Commands

Initial Setup

pip install -r requirements.txt

Create .env file with:

GEMINI_API_KEY="your_gemini_api_key_here"
FLASK_SECRET_KEY="optional_custom_secret_key"

Preprocessing (Only when updating party programs)

python preprocess_programs.py

Converts PDF/Markdown party programs to JSON cache. The cache is included in git repository for fast deployments. Only run this when adding/updating party programs, then commit the updated cache.

Running the Application

python app.py

Starts the Flask web server on port 8080. Will use cached programs if available, fallback to direct loading if not.

Testing

python main.py

Runs analysis with predefined answers against party programs using Gemini API.

Architecture

Core Application Structure

  • app.py: Main Flask application with two distinct modes - political quiz and Q&A chat
  • utils.py: File processing utilities for loading PDF and Markdown party programs
  • main.py: Testing script for analyzing predefined answers against party programs

Party Program System

The application uses a two-tier loading system:

  1. Pre-built Cache: JSON cache in ./cache/ directory is included in git repository (6MB) for instant deployments
  2. Lazy Loading: PartyProgramCache class loads programs on-demand from JSON cache, with fallback to direct file loading

The cache includes topic extraction for faster content retrieval. When party programs are updated, run preprocess_programs.py locally and commit the updated cache.

Dual-Mode Architecture

  1. Quiz Mode: 5-question political stance assessment with session management

    • Generates dynamic questions using Gemini
    • Tracks user answers in Flask sessions
    • Analyzes political stance and matches to best-fitting party

  2. Chat Mode: Direct Q&A about specific parties or general political topics

    • Detects party names from user messages using comprehensive mapping system
    • Extracts relevant content sections (max 20k chars) based on question keywords
    • Supports both single-party and multi-party comparisons

Content Optimization

The system includes intelligent content extraction that:

  • Maps political topics to relevant keywords (taxes, immigration, environment, etc.)
  • Extracts only relevant paragraphs for efficient LLM processing
  • Limits content to 20,000 characters per party to manage API costs

Party Detection Logic

Complex fuzzy matching system in detect_parties_from_message() handles:

  • Norwegian party name variations and abbreviations
  • Word boundary matching to avoid false positives
  • Longest-match-first algorithm for accurate detection
  • Special handling for generic terms like "partiene" (the parties)

Session Management

Quiz functionality uses Flask sessions to maintain state across:

  • User answers collection
  • Question generation tracking
  • Quiz completion status
  • Temporary session cleanup after analysis