"""
Pattern 009: Self-Refine
Description:
Iteratively improves outputs through self-critique and refinement.
The agent generates an initial output, evaluates it, identifies weaknesses,
and produces an improved version. This cycle repeats until quality threshold
is met or max iterations reached.
Key Concepts:
- Initial Generation: Create first draft output
- Self-Critique: Analyze weaknesses and areas for improvement
- Refinement: Generate improved version based on critique
- Iteration: Repeat until satisfactory or max iterations
- Quality Tracking: Monitor improvement across iterations
Benefits:
- Higher quality outputs through iteration
- Self-improvement without external feedback
- Transparency in refinement process
- Handles complex generation tasks
Use Cases:
- Content creation (writing, code, reports)
- Problem-solving with iterative improvement
- Design tasks requiring multiple revisions
- Any task benefiting from refinement
LangChain Implementation:
Chained prompts for generation, critique, and refinement with iteration control.
"""
import os
from typing import List, Dict, Any, Optional
from dataclasses import dataclass, field
from enum import Enum
from dotenv import load_dotenv
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_openai import ChatOpenAI
# Load environment variables
load_dotenv()
class IterationStatus(Enum):
"""Status of refinement iteration."""
INITIAL = "initial"
REFINING = "refining"
COMPLETED = "completed"
MAX_ITERATIONS = "max_iterations_reached"
@dataclass
class Critique:
"""Represents a critique of generated output."""
strengths: List[str] = field(default_factory=list)
weaknesses: List[str] = field(default_factory=list)
suggestions: List[str] = field(default_factory=list)
score: float = 0.0 # Quality score 0-1
@dataclass
class RefinementIteration:
"""Represents one iteration of the refinement process."""
iteration: int
output: str
critique: Optional[Critique] = None
status: IterationStatus = IterationStatus.INITIAL
def __str__(self):
return f"Iteration {self.iteration}: {self.status.value} (score: {self.critique.score if self.critique else 'N/A'})"
class SelfRefineAgent:
"""Agent that uses self-refinement to improve outputs."""
def __init__(self, model_name: str = "gpt-3.5-turbo",
max_iterations: int = 3,
quality_threshold: float = 0.8):
"""
Initialize the Self-Refine agent.
Args:
model_name: Name of the OpenAI model
max_iterations: Maximum refinement iterations
quality_threshold: Stop if quality score exceeds this
"""
self.llm = ChatOpenAI(
model=model_name,
temperature=0.7,
api_key=os.getenv("OPENAI_API_KEY")
)
self.critic_llm = ChatOpenAI(
model=model_name,
temperature=0.3, # Lower temperature for more consistent critique
api_key=os.getenv("OPENAI_API_KEY")
)
self.max_iterations = max_iterations
self.quality_threshold = quality_threshold
self.history: List[RefinementIteration] = []
def generate_initial(self, task: str, context: str = "") -> str:
"""
Generate initial output for the task.
Args:
task: The task description
context: Additional context
Returns:
Initial generated output
"""
prompt = ChatPromptTemplate.from_messages([
("system", """Generate a response to the given task.
Focus on creating a solid first draft. It doesn't need to be perfect yet."""),
("human", """{context}
Task: {task}
Response:""")
])
chain = prompt | self.llm | StrOutputParser()
output = chain.invoke({
"task": task,
"context": context if context else "No additional context."
})
return output.strip()
def critique(self, task: str, output: str, iteration: int) -> Critique:
"""
Generate critique of current output.
Args:
task: Original task
output: Current output to critique
iteration: Current iteration number
Returns:
Critique object with feedback
"""
prompt = ChatPromptTemplate.from_messages([
("system", """You are a constructive critic. Evaluate the response to the task.
Provide:
1. Strengths: What works well (2-3 points)
2. Weaknesses: What could be improved (2-3 points)
3. Suggestions: Specific improvements to make (2-3 points)
4. Score: Overall quality score from 0.0 to 1.0
Format your response as:
STRENGTHS:
- [strength 1]
- [strength 2]
WEAKNESSES:
- [weakness 1]
- [weakness 2]
SUGGESTIONS:
- [suggestion 1]
- [suggestion 2]
SCORE: [0.0-1.0]"""),
("human", """Task: {task}
Current Response (Iteration {iteration}):
{output}
Critique:""")
])
chain = prompt | self.critic_llm | StrOutputParser()
critique_text = chain.invoke({
"task": task,
"output": output,
"iteration": iteration
})
# Parse critique
critique = Critique()
sections = {
"strengths": [],
"weaknesses": [],
"suggestions": []
}
current_section = None
for line in critique_text.split('\n'):
line = line.strip()
if line.upper().startswith('STRENGTHS:'):
current_section = "strengths"
elif line.upper().startswith('WEAKNESSES:'):
current_section = "weaknesses"
elif line.upper().startswith('SUGGESTIONS:'):
current_section = "suggestions"
elif line.upper().startswith('SCORE:'):
try:
score_str = line.split(':', 1)[1].strip()
critique.score = float(score_str)
except:
critique.score = 0.5
elif line.startswith('-') and current_section:
sections[current_section].append(line.lstrip('- ').strip())
critique.strengths = sections["strengths"]
critique.weaknesses = sections["weaknesses"]
critique.suggestions = sections["suggestions"]
# Ensure score is set
if critique.score == 0.0 and (critique.strengths or not critique.weaknesses):
critique.score = 0.6 # Default reasonable score
return critique
def refine(self, task: str, current_output: str, critique: Critique) -> str:
"""
Generate refined version based on critique.
Args:
task: Original task
current_output: Current output to refine
critique: Critique with improvement suggestions
Returns:
Refined output
"""
weaknesses_text = "\n".join([f"- {w}" for w in critique.weaknesses])
suggestions_text = "\n".join([f"- {s}" for s in critique.suggestions])
prompt = ChatPromptTemplate.from_messages([
("system", """Improve the response based on the critique provided.
Address the weaknesses and implement the suggestions while maintaining the strengths.
Produce a refined, higher-quality version."""),
("human", """Task: {task}
Current Response:
{current_output}
Weaknesses to Address:
{weaknesses}
Suggestions to Implement:
{suggestions}
Refined Response:""")
])
chain = prompt | self.llm | StrOutputParser()
refined = chain.invoke({
"task": task,
"current_output": current_output,
"weaknesses": weaknesses_text,
"suggestions": suggestions_text
})
return refined.strip()
def refine_iteratively(self, task: str, context: str = "",
verbose: bool = True) -> Dict[str, Any]:
"""
Perform iterative refinement on the task.
Args:
task: The task to complete
context: Additional context
verbose: Whether to print progress
Returns:
Dictionary with final output and refinement history
"""
if verbose:
print(f"\nTask: {task}\n")
print("="*60)
print("ITERATION 1: Initial Generation")
print("="*60)
# Generate initial output
initial_output = self.generate_initial(task, context)
iteration = RefinementIteration(
iteration=1,
output=initial_output,
status=IterationStatus.INITIAL
)
self.history.append(iteration)
if verbose:
print(f"\nGenerated:\n{initial_output[:200]}..." if len(initial_output) > 200 else f"\nGenerated:\n{initial_output}")
current_output = initial_output
# Refinement loop
for i in range(2, self.max_iterations + 2):
if verbose:
print(f"\n{'='*60}")
print(f"ITERATION {i}: Critique & Refine")
print("="*60)
# Critique current output
critique = self.critique(task, current_output, i - 1)
# Update previous iteration with critique
self.history[-1].critique = critique
if verbose:
print(f"\nCritique:")
print(f" Quality Score: {critique.score:.2f}")
if critique.strengths:
print(f" Strengths: {len(critique.strengths)} identified")
if critique.weaknesses:
print(f" Weaknesses: {len(critique.weaknesses)} identified")
if critique.suggestions:
print(f" Suggestions: {len(critique.suggestions)} provided")
# Check if quality threshold met
if critique.score >= self.quality_threshold:
self.history[-1].status = IterationStatus.COMPLETED
if verbose:
print(f"\n✓ Quality threshold ({self.quality_threshold}) met!")
break
# Check max iterations
if i > self.max_iterations:
self.history[-1].status = IterationStatus.MAX_ITERATIONS
if verbose:
print(f"\n⚠ Max iterations ({self.max_iterations}) reached")
break
# Refine output
if verbose:
print(f"\nRefining...")
refined_output = self.refine(task, current_output, critique)
iteration = RefinementIteration(
iteration=i,
output=refined_output,
status=IterationStatus.REFINING
)
self.history.append(iteration)
if verbose:
print(f"\nRefined:\n{refined_output[:200]}..." if len(refined_output) > 200 else f"\nRefined:\n{refined_output}")
current_output = refined_output
# Final critique
if self.history[-1].critique is None:
final_critique = self.critique(task, current_output, len(self.history))
self.history[-1].critique = final_critique
return {
"task": task,
"final_output": current_output,
"iterations": len(self.history),
"final_score": self.history[-1].critique.score if self.history[-1].critique else 0.0,
"history": self.history,
"status": self.history[-1].status.value
}
def get_improvement_summary(self) -> str:
"""Generate summary of improvements across iterations."""
if len(self.history) < 2:
return "Not enough iterations to compare."
initial_score = self.history[0].critique.score if self.history[0].critique else 0.0
final_score = self.history[-1].critique.score if self.history[-1].critique else 0.0
improvement = final_score - initial_score
improvement_pct = (improvement / max(initial_score, 0.1)) * 100
summary = f"""
Improvement Summary:
Initial Score: {initial_score:.2f}
Final Score: {final_score:.2f}
Improvement: {improvement:+.2f} ({improvement_pct:+.1f}%)
Total Iterations: {len(self.history)}
"""
return summary
def demonstrate_self_refine():
"""Demonstrates the Self-Refine pattern."""
print("=" * 80)
print("PATTERN 009: Self-Refine")
print("=" * 80)
print()
print("Self-Refine iteratively improves outputs through:")
print("1. Initial Generation: Create first draft")
print("2. Self-Critique: Analyze strengths and weaknesses")
print("3. Refinement: Generate improved version")
print("4. Iteration: Repeat until quality threshold met")
print()
# Create agent
agent = SelfRefineAgent(max_iterations=3, quality_threshold=0.85)
# Test tasks
tasks = [
{
"task": "Write a brief explanation of quantum entanglement for a general audience",
"context": "Make it accessible but accurate, around 100 words"
},
{
"task": "Create a function to calculate the Fibonacci sequence efficiently",
"context": "Use Python, include docstring and handle edge cases"
}
]
for idx, test_case in enumerate(tasks, 1):
print(f"\n{'='*80}")
print(f"Example {idx}")
print('='*80)
try:
# Reset history
agent.history = []
result = agent.refine_iteratively(
task=test_case['task'],
context=test_case['context'],
verbose=True
)
print(f"\n\n{'='*80}")
print("FINAL RESULT")
print('='*80)
print(f"\nFinal Output:\n{result['final_output']}")
print(f"\n{agent.get_improvement_summary()}")
print(f"Status: {result['status']}")
except Exception as e:
print(f"\n✗ Error: {str(e)}")
import traceback
traceback.print_exc()
# Summary
print("\n\n" + "=" * 80)
print("SELF-REFINE PATTERN DEMONSTRATION COMPLETE")
print("=" * 80)
print()
print("Key Features Demonstrated:")
print("1. Iterative Improvement: Multiple refinement cycles")
print("2. Self-Critique: Autonomous evaluation of outputs")
print("3. Targeted Refinement: Addressing specific weaknesses")
print("4. Quality Tracking: Score-based improvement monitoring")
print("5. Convergence: Stops when threshold met or max iterations")
print()
print("Advantages:")
print("- Higher quality outputs through iteration")
print("- No external feedback required")
print("- Transparent improvement process")
print("- Handles complex generation tasks")
print()
print("When to use Self-Refine:")
print("- Content creation (writing, code, reports)")
print("- Tasks benefiting from multiple revisions")
print("- When quality is more important than speed")
print("- Complex generation requiring refinement")
print()
print("LangChain Components Used:")
print("- ChatPromptTemplate: Generation, critique, and refinement prompts")
print("- StrOutputParser: Parse LLM outputs")
print("- Separate LLMs: Different temperatures for generation vs critique")
print("- Iteration control: Quality threshold and max iterations")
print()
if __name__ == "__main__":
# Check for API key
if not os.getenv("OPENAI_API_KEY"):
print("Error: OPENAI_API_KEY not found in environment variables.")
print("Please set it in your .env file or environment.")
exit(1)
demonstrate_self_refine()
Agentic Design Pattern: Self Refine
Pattern 009: Self Refine – Line-by-Line Explanation
Overview
This document provides a detailed line-by-line explanation of the Self Refine pattern implementation.
Pattern Description
Pattern 009: Self-Refine
Description: Iteratively improves outputs through self-critique and refinement. The agent generates an initial output, evaluates it, identifies weaknesses, and produces an improved version. This cycle repeats until quality threshold is met or max iterations reached.
Key Concepts: – Initial Generation: Create first draft output – Self-Critique: Analyze weaknesses and areas for improvement – Refinement: Generate improved version based on critique – Iteration: Repeat until satisfactory or max iterations – Quality Tracking: Monitor improvement across iterations
Benefits: – Higher quality outputs through iteration – Self-improvement without external feedback – Transparency in refinement process – Handles complex generation tasks
Use Cases: – Content creation (writing, code, reports) – Problem-solving with iterative improvement – Design tasks requiring multiple revisions – Any task benefiting from refinement
LangChain Implementation: Chained prompts for generation, critique, and refinement with iteration control.
File Statistics
- Total Lines: 486
- Classes: 4
- Functions: 1
- Constants: 0
Module Documentation
"""
Pattern 009: Self-Refine
Description:
Iteratively improves outputs through self-critique and refinement.
The agent generates an initial output, evaluates it, identifies weaknesses,
and produces an improved version. This cycle repeats until quality threshold
is met or max iterations reached.
Key Concepts:
- Initial Generation: Create first draft output
- Self-Critique: Analyze weaknesses and areas for improvement
- Refinement: Generate improved version based on critique
- Iteration: Repeat until satisfactory or max iterations
- Quality Tracking: Monitor improvement across iterations
Benefits:
- Higher quality outputs through iteration
- Self-improvement without external feedback
- Transparency in refinement process
- Handles complex generation tasks
Use Cases:
- Content creation (writing, code, reports)
- Problem-solving with iterative improvement
- Design tasks requiring multiple revisions
- Any task benefiting from refinement
LangChain Implementation:
Chained prompts for generation, critique, and refinement with iteration control.
"""
Purpose: This module implements the Self Refine pattern, providing a structured approach to agent design and implementation.
Import Statements
Standard Library Imports
import dataclasses.dataclass
import dataclasses.field
import enum.Enum
import langchain_core.output_parsers.StrOutputParser
import langchain_core.prompts.ChatPromptTemplate
import os
import typing.Any
import typing.Dict
import typing.List
import typing.Optional
Purpose: Core Python functionality:
– dataclasses.dataclass: Simplified class definitions with @dataclass
– dataclasses.field: Simplified class definitions with @dataclass
– enum.Enum: Enumeration support
– langchain_core.output_parsers.StrOutputParser: Regular expression operations
– langchain_core.prompts.ChatPromptTemplate: Regular expression operations
– os: Environment variable access and system operations
– typing.Any: Type hints for better code documentation (List, Dict, Any, Optional)
– typing.Dict: Type hints for better code documentation (List, Dict, Any, Optional)
– typing.List: Type hints for better code documentation (List, Dict, Any, Optional)
– typing.Optional: Type hints for better code documentation (List, Dict, Any, Optional)
LangChain Imports
from langchain_openai.ChatOpenAI
Purpose: LangChain framework components: – LLM Interface: Wrapper for OpenAI models – Chains: Composable building blocks for complex workflows
Other Imports
from dotenv.load_dotenv
from traceback
Purpose: Additional dependencies for specialized functionality.
Classes
Class: IterationStatus
Lines 47-52
Inherits from: Enum
Purpose:
Status of refinement iteration.
Class: Critique
Lines 56-61
Purpose:
Represents a critique of generated output.
Class: RefinementIteration
Lines 65-73
Purpose:
Represents one iteration of the refinement process.
Methods:
__str__(self)
Lines 72-73
Purpose: Str operation.
Parameters:
Class: SelfRefineAgent
Lines 76-387
Purpose:
Agent that uses self-refinement to improve outputs.
Methods:
__init__(self, model_name, max_iterations, quality_threshold)
Lines 79-104
Purpose:
Initialize the Self-Refine agent.
Args:
model_name: Name of the OpenAI model
max_iterations: Maximum refinement iterations
quality_threshold: Stop if quality score exceeds this
Parameters:
– model_name: Language model instance
– max_iterations: Max iterations
– quality_threshold: Quality threshold
generate_initial(self, task, context)
Lines 106-133
Purpose:
Generate initial output for the task.
Args:
task: The task description
context: Additional context
Returns:
Initial generated output
Parameters:
– task: Task
– context: Context
Returns: str
critique(self, task, output, iteration)
Lines 135-221
Purpose:
Generate critique of current output.
Args:
task: Original task
output: Current output to critique
iteration: Current iteration number
Returns:
Critique object with feedback
Parameters:
– task: Task
– output: Output
– iteration: Iteration
Returns: Critique
refine(self, task, current_output, critique)
Lines 223-264
Purpose:
Generate refined version based on critique.
Args:
task: Original task
current_output: Current output to refine
critique: Critique with improvement suggestions
Returns:
Refined output
Parameters:
– task: Task
– current_output: Current output
– critique: Critique
Returns: str
refine_iteratively(self, task, context, verbose)
Lines 266-367
Purpose:
Perform iterative refinement on the task.
Args:
task: The task to complete
context: Additional context
verbose: Whether to print progress
Returns:
Dictionary with final output and refinement history
Parameters:
– task: Task
– context: Context
– verbose: Enable detailed logging
Returns: <ast.Subscript object at 0x000001A06D658D90>
get_improvement_summary(self)
Lines 369-387
Purpose:
Generate summary of improvements across iterations.
Parameters:
Returns: str
Functions
demonstrate_self_refine()
Lines 390-475
Purpose:
Demonstrates the Self-Refine pattern.
Key Concepts Summary
Pattern Overview
The Self Refine pattern provides a structured approach to building AI agents with specific capabilities and behaviors. This implementation leverages LangChain’s composable components to create a robust and maintainable solution.
Core Components
- Agent Class: Main implementation of the pattern logic
- Prompt Engineering: Carefully crafted prompts to guide LLM behavior
- Tool Integration: Optional tools for extended functionality
- Memory Management: State handling across interactions
- Output Parsing: Structured response processing
When to Use
✅ Good for: – Scenarios requiring this specific pattern’s strengths – Applications needing structured agent behavior – Systems where explainability is important – Production environments requiring reliability
❌ Not ideal for: – Simple single-prompt tasks – Scenarios requiring different pattern characteristics – Resource-constrained environments (if computationally intensive)
Best Practices Demonstrated
1. Environment Configuration
- ✅ Use
.envfiles for API keys - ✅ Never hardcode credentials
- ✅ Validate environment variables on startup
2. Error Handling
- ✅ Graceful degradation on failures
- ✅ Informative error messages
- ✅ Logging for debugging
3. Prompt Engineering
- ✅ Clear, specific instructions
- ✅ Structured output formats
- ✅ Few-shot examples when beneficial
4. Code Organization
- ✅ Separation of concerns (class-based design)
- ✅ Type hints for clarity
- ✅ Docstrings for documentation
5. Performance Optimization
- ✅ Appropriate temperature settings
- ✅ Token budget management
- ✅ Caching where applicable
LangChain Components Used
- ChatOpenAI: OpenAI model interface
- Chains: Sequential component composition
Why LangChain? – Composable abstractions for LLM applications – Rich ecosystem of integrations – Production-ready patterns – Active community and development
Execution Flow
1. Initialize agent with configuration
↓
2. Load environment and dependencies
↓
3. Prepare prompt with input data
↓
4. Invoke LLM through LangChain
↓
5. Parse and validate response
↓
6. Return formatted result
Key Decision Points: – Model selection (temperature, model type) – Prompt structure (few-shot vs zero-shot) – Error handling strategy – Output format validation
Performance Considerations
Computational Cost
- LLM Calls: Primary cost factor (based on tokens)
- Latency: Depends on model and input size
- Memory: Minimal for most patterns
Optimization Strategies
- Prompt Optimization: Minimize tokens while maintaining clarity
- Caching: Cache repeated queries/responses
- Model Selection: Use smaller models when appropriate
- Batch Processing: Group similar requests
- Streaming: For long responses, stream tokens
Cost Estimates (Approximate)
- GPT-3.5-turbo: ~$0.002 per request
- GPT-4: ~$0.03 per request
- GPT-4-turbo: ~$0.01 per request
Actual costs vary based on input/output token counts
Common Pitfalls to Avoid
1. Prompt Design Issues
❌ Vague or ambiguous instructions
✅ Clear, specific, testable prompts
2. Error Handling
❌ Ignoring API errors or rate limits
✅ Implement retry logic and graceful degradation
3. Token Management
❌ Exceeding model token limits
✅ Truncate inputs, chunk processing
4. Security
❌ Exposing API keys in code
✅ Use environment variables
5. Testing
❌ Only testing happy paths
✅ Test edge cases, failures, malformed inputs
6. Cost Management
❌ Unmonitored API usage
✅ Implement budgets, alerts, and monitoring
Further Improvements
Production Enhancements
- Monitoring: Add logging, metrics, tracing
- Validation: Input/output schema validation
- Rate Limiting: Implement request throttling
- Caching: Redis or in-memory cache for responses
- Async Support: Non-blocking operations for scalability
Feature Extensions
- Multi-Model Support: Switch between providers
- Fallback Models: Graceful degradation strategy
- Custom Tools: Domain-specific function calling
- Advanced Memory: Vector stores for long-term memory
- Evaluation: Automated testing and benchmarking
Code Quality
- Testing: Unit tests, integration tests
- Documentation: API docs, usage examples
- Type Safety: Stricter type hints
- Configuration: Externalized config files
Related Patterns
- Pattern 008
- Pattern 010
- Other Reasoning Patterns (001-011)
Pattern Families: – Reasoning Patterns (001-011) – Multi-Agent Patterns (012-019) – Tool Use Patterns (020-025) – Memory Patterns (026-031) – And many more…
Summary
The Self Refine pattern provides a powerful approach to building AI agents with specific capabilities. This implementation demonstrates:
✅ Key Strengths: – Clean, maintainable code structure – LangChain integration for composability – Type hints and documentation – Production-ready patterns
🎯 Best Use Cases: – Applications requiring this pattern’s characteristics – Systems needing structured agent behavior – Production environments requiring reliability
📚 Learning Resources: – LangChain Documentation: https://python.langchain.com/ – Pattern Documentation: See related patterns – OpenAI API Reference: https://platform.openai.com/docs/
File: 009_self_refine.py
Total Lines: 486
Pattern Category: AI Agent Design Pattern #009
Generated from actual implementation code
For educational and reference purposes
Leave a Reply