Skip to main content
A feedback loop workflow where LLM-generated outputs are evaluated, refined, and optimized iteratively to improve accuracy and relevance.

​
Quick Start

1

Install Package

First, install the PraisonAI Agents package:
pip install praisonaiagents
2

Set API Key

Set your OpenAI API key as an environment variable in your terminal:
export OPENAI_API_KEY=your_api_key_here
3

Create a file

Create a new file app.py with the basic setup:
from praisonaiagents import Agent, AgentFlow, Task
from praisonaiagents import repeat

# Create generator agent
generator = Agent(
    name="Generator",
    role="Content Generator",
    goal="Generate high-quality content",
    instructions="Generate content based on the topic. Improve based on feedback if provided."
)

# Create evaluator agent  
evaluator = Agent(
    name="Evaluator",
    role="Content Evaluator",
    goal="Evaluate and provide feedback on content",
    instructions="Evaluate the content quality. If good, respond with 'APPROVED'. Otherwise provide specific feedback for improvement."
)

# Evaluation function to check if content is approved
def is_approved(ctx) -> bool:
    return "approved" in ctx.previous_result.lower()

# Create workflow with evaluator-optimizer pattern
workflow = AgentFlow(
    steps=[
        generator,  # First, generate content
        repeat(
            evaluator,  # Evaluate and provide feedback
            until=is_approved,
            max_iterations=3
        )
    ]
)

# Run optimization workflow
result = workflow.start("Write a compelling product description for an AI assistant")
print(f"Final Result: {result['output'][:500]}...")
4

Start Workflow

Type this in your terminal to run your workflow:
python app.py
Requirements
  • Python 3.10 or higher
  • OpenAI API key. Generate OpenAI API key here. Use Other models using this guide.
  • Basic understanding of Python

​
Understanding Evaluator-Optimizer

What is Evaluator-Optimizer?

Evaluator-Optimizer pattern enables:
  • Iterative solution generation and refinement
  • Automated quality evaluation
  • Feedback-driven optimization
  • Continuous improvement loops

​
Features

Solution Generation

Generate solutions based on requirements and feedback.

Quality Evaluation

Automatically assess solution quality and completeness.

Feedback Loop

Implement iterative improvement through feedback cycles.

Process Control

Monitor and control the optimization process.

​
Configuration Options

# Create a generator agent
generator = Agent(
    name="Generator",
    role="Solution generator",
    goal="Generate and improve solutions",
    instructions="Step-by-step instructions for generation",
      # Enable detailed logging
)

# Create an evaluator agent
evaluator = Agent(
    name="Evaluator",
    role="Solution evaluator",
    goal="Evaluate and provide feedback",
    instructions="Evaluation criteria and feedback format"
)

# Create tasks with feedback loop
generate_task = Task(
    name="generate",
    description="Generate solution",
    agent=generator,
    is_start=True,
    task_type="decision",
    next_tasks=["evaluate"]
)

evaluate_task = Task(
    name="evaluate",
    description="Evaluate solution",
    agent=evaluator,
    context=[generate_task],
    task_type="decision",
    condition={
        "more": ["generate"],  # Continue optimization
        "done": [""]  # Exit when complete
    }
)

​
Troubleshooting

Generation Issues

If generation is not improving:
  • Review generator instructions
  • Check feedback integration
  • Enable verbose mode for debugging

Evaluation Flow

If evaluation cycle is incorrect:
  • Verify evaluation criteria
  • Check condition mappings
  • Review feedback loop connections

​
Next Steps

AutoAgents

Learn about automatically created and managed AI agents

Mini Agents

Explore lightweight, focused AI agents
For optimal results, ensure your generator instructions and evaluation criteria are clear and well-defined to achieve the desired optimization outcomes.