LLM Writer-Critic System

An agentic joke generation application powered by Groq's LLaMA model, featuring a dual-agent architecture with independent writer and critic roles for quality-controlled humor generation.

Overview

This project implements a sophisticated Writer-Critic pattern for generating high-quality jokes. The system uses two specialized LLM agents:

• Writer Agent: Generates creative, original jokes with high creativity settings (temperature: 0.9)
• Critic Agent: Reviews jokes against quality criteria with strict, deterministic evaluation (temperature: 0.0)

graph TD
    A[User Request] --> B[Writer Agent]
    B -->|Generate Joke| C[Critic Agent]
    C -->|Approved| D[Display Joke]
    C -->|Rejected| E[Retry Loop]
    E -->|Max Retries| F[Fail]
    E -->|Retry| B
    
    B -.->|Temperature 0.9| G[High Creativity]
    C -.->|Temperature 0.0| H[Strict Evaluation]

Loading

Features

• Multi-category jokes: programming, dad jokes, dark humor, one-liners, slapstick, knock-knock, tech, career, relationships, random
• Multi-language support: English, Spanish, French, German, Italian, Portuguese, Hindi, Japanese, Chinese, Korean, Russian, Arabic, Tamil, Telugu
• Quality control: Automatic review against humor, appropriateness, category match, originality, and language criteria
• Retry mechanism: Up to 3 attempts per joke with feedback integration
• Interactive CLI: User-friendly menu for category/language selection

Project Structure

llm-writer-critic-system/
├── README.md              # This file
├── requirements.txt       # Python dependencies
├── config.yaml            # Configuration settings
├── .env.example           # Environment template
└── app/
    ├── __init__.py
    ├── main.py            # Main CLI application
    ├── llm.py             # LLM initialization & validation
    ├── writer.py          # Writer agent for joke generation
    └── critic.py          # Critic agent for quality review

Installation

1. Clone the repository

   cd llm-writer-critic-system

2. Create virtual environment

   python -m venv venv
   source venv/bin/activate  # Linux/Mac
   # or
   .\venv\Scripts\activate   # Windows

3. Install dependencies

   pip install -r requirements.txt

4. Configure environment

   cp .env.example .env
   # Edit .env and add your Groq API key

Configuration

Edit config.yaml to customize:

• LLM provider and model settings
• Temperature parameters (writer creativity vs. critic strictness)
• Application defaults (max retries, categories, languages)
• Logging preferences

Usage

python -m app.main

Controls

• n - Generate next joke
• c - Change category
• l - Change language
• m - Show full menu
• e - Exit

Architecture Details

Writer Agent (app/writer.py)

Generates jokes using a carefully crafted prompt with:

• Category-specific descriptions
• Language targeting
• Previous attempt avoidance
• Creative temperature (0.9)

Critic Agent (app/critic.py)

Evaluates jokes against six criteria:

1. Funny - Genuine humor
2. Appropriate - Clean and suitable
3. Category Match - Fits the theme
4. Original - Creative, not clichéd
5. Concise - Short (1-3 lines)
6. Language - Correctly written

LLM Interface (app/llm.py)

Uses LangChain's ChatGroq integration with Groq's llama-3.1-8b-instant model.

Requirements

• Python 3.8+
• Groq API key (get one here)
• Dependencies: langchain-groq, python-dotenv, langchain-core

License

MIT License