"""
Pattern 169: Retrieval Interleaving
Description:
The Retrieval Interleaving pattern dynamically retrieves information during
generation, interleaving retrieval steps with reasoning and generation steps.
Unlike standard RAG which retrieves once upfront, this pattern retrieves
multiple times as needed based on the generation process.
Components:
1. Generation Monitor: Tracks generation progress
2. Retrieval Trigger: Decides when to retrieve
3. Query Generator: Creates retrieval queries
4. Retriever: Fetches relevant information
5. Context Integrator: Merges retrieved content
6. Iteration Controller: Manages interleaving loops
Use Cases:
- Complex multi-hop reasoning
- Long-form content generation
- Research-intensive tasks
- Dynamic information needs
- Iterative question answering
- Knowledge synthesis
Benefits:
- Adaptive information gathering
- Better context relevance
- Handles complex queries
- Reduces initial retrieval overhead
- More accurate results
Trade-offs:
- Multiple retrieval calls
- Increased latency
- Complex orchestration
- Higher costs
- Requires good triggering logic
LangChain Implementation:
Implements iterative retrieval with LangChain's retrieval components.
Uses LLM to decide when and what to retrieve during generation.
"""
import os
from typing import List, Dict, Any, Optional, Tuple
from dataclasses import dataclass, field
from enum import Enum
from dotenv import load_dotenv
from langchain_openai import ChatOpenAI, OpenAIEmbeddings
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.documents import Document
load_dotenv()
class RetrievalDecision(Enum):
"""Decision on whether to retrieve"""
RETRIEVE = "retrieve"
CONTINUE_GENERATION = "continue"
COMPLETE = "complete"
@dataclass
class RetrievalStep:
"""Represents a retrieval step"""
query: str
documents: List[Document]
timestamp: int
reasoning: str
@dataclass
class GenerationStep:
"""Represents a generation step"""
content: str
timestamp: int
used_retrieval: bool
@dataclass
class InterleavedResult:
"""Result of interleaved retrieval-generation"""
final_output: str
retrieval_steps: List[RetrievalStep]
generation_steps: List[GenerationStep]
total_retrievals: int
total_tokens_estimated: int
class MockVectorStore:
"""Mock vector store for demonstration"""
def __init__(self):
"""Initialize with sample documents"""
self.documents = [
Document(page_content="Machine learning is a subset of AI that enables computers to learn from data.",
metadata={"source": "ml_basics"}),
Document(page_content="Deep learning uses neural networks with multiple layers to process data.",
metadata={"source": "dl_intro"}),
Document(page_content="Natural language processing (NLP) enables computers to understand human language.",
metadata={"source": "nlp_overview"}),
Document(page_content="Transformers revolutionized NLP with attention mechanisms.",
metadata={"source": "transformers"}),
Document(page_content="GPT models are large language models trained on vast amounts of text data.",
metadata={"source": "gpt_info"}),
Document(page_content="RAG combines retrieval with generation for better accuracy.",
metadata={"source": "rag_concept"}),
Document(page_content="Computer vision enables machines to interpret visual information.",
metadata={"source": "cv_basics"}),
Document(page_content="Reinforcement learning trains agents through rewards and penalties.",
metadata={"source": "rl_intro"}),
]
self.embeddings = OpenAIEmbeddings()
def similarity_search(self, query: str, k: int = 3) -> List[Document]:
"""Mock similarity search"""
# Simple keyword matching for demonstration
query_lower = query.lower()
scored_docs = []
for doc in self.documents:
content_lower = doc.page_content.lower()
score = sum(1 for word in query_lower.split() if word in content_lower)
scored_docs.append((score, doc))
scored_docs.sort(key=lambda x: x[0], reverse=True)
return [doc for _, doc in scored_docs[:k]]
class RetrievalInterleavingAgent:
"""Agent that interleaves retrieval with generation"""
def __init__(self, vector_store: Optional[MockVectorStore] = None,
max_retrievals: int = 5):
"""
Initialize interleaving agent
Args:
vector_store: Vector store for retrieval
max_retrievals: Maximum number of retrieval steps
"""
self.vector_store = vector_store or MockVectorStore()
self.max_retrievals = max_retrievals
self.llm = ChatOpenAI(model="gpt-3.5-turbo", temperature=0.7)
# Prompts
self.decision_prompt = ChatPromptTemplate.from_messages([
("system", """You are an AI that decides when to retrieve more information.
Analyze the current generation state and decide if you need more information.
Respond with one of: RETRIEVE, CONTINUE, COMPLETE
If RETRIEVE, also provide a search query."""),
("user", """Query: {query}
Current generation:
{current_generation}
Context so far:
{context}
Decision (RETRIEVE/CONTINUE/COMPLETE) and query if retrieving:""")
])
self.generation_prompt = ChatPromptTemplate.from_messages([
("system", """You are a helpful AI assistant. Generate content based on
the query and available context. Be factual and cite information when possible."""),
("user", """Query: {query}
Available context:
{context}
Previous generation:
{previous_generation}
Continue or complete the response:""")
])
self.decision_chain = self.decision_prompt | self.llm | StrOutputParser()
self.generation_chain = self.generation_prompt | self.llm | StrOutputParser()
def generate_with_interleaved_retrieval(self, query: str) -> InterleavedResult:
"""Generate response with interleaved retrieval"""
retrieval_steps: List[RetrievalStep] = []
generation_steps: List[GenerationStep] = []
current_generation = ""
retrieved_context = ""
step_counter = 0
# Initial generation attempt
initial_gen = self._generate(query, "", "")
current_generation = initial_gen
generation_steps.append(GenerationStep(
content=initial_gen,
timestamp=step_counter,
used_retrieval=False
))
step_counter += 1
# Interleaving loop
for retrieval_count in range(self.max_retrievals):
# Decide if we need more information
decision, retrieval_query = self._decide_next_step(
query, current_generation, retrieved_context
)
if decision == RetrievalDecision.COMPLETE:
break
if decision == RetrievalDecision.RETRIEVE and retrieval_query:
# Retrieve documents
docs = self.vector_store.similarity_search(retrieval_query, k=2)
retrieval_steps.append(RetrievalStep(
query=retrieval_query,
documents=docs,
timestamp=step_counter,
reasoning=f"Needed information about: {retrieval_query}"
))
step_counter += 1
# Update context
new_context = "\n\n".join([doc.page_content for doc in docs])
retrieved_context += "\n\n" + new_context
# Generate with new context
new_generation = self._generate(query, retrieved_context, current_generation)
current_generation = new_generation
generation_steps.append(GenerationStep(
content=new_generation,
timestamp=step_counter,
used_retrieval=True
))
step_counter += 1
elif decision == RetrievalDecision.CONTINUE_GENERATION:
# Continue generation without retrieval
new_generation = self._generate(query, retrieved_context, current_generation)
current_generation = new_generation
generation_steps.append(GenerationStep(
content=new_generation,
timestamp=step_counter,
used_retrieval=False
))
step_counter += 1
# Estimate tokens
total_tokens = sum(len(step.content.split()) for step in generation_steps) * 1.3
total_tokens += sum(len(step.query.split()) for step in retrieval_steps) * 1.3
return InterleavedResult(
final_output=current_generation,
retrieval_steps=retrieval_steps,
generation_steps=generation_steps,
total_retrievals=len(retrieval_steps),
total_tokens_estimated=int(total_tokens)
)
def _decide_next_step(self, query: str, current_generation: str,
context: str) -> Tuple[RetrievalDecision, Optional[str]]:
"""Decide whether to retrieve, continue, or complete"""
decision_text = self.decision_chain.invoke({
"query": query,
"current_generation": current_generation[:200],
"context": context[:300] if context else "No context yet"
})
# Parse decision
decision_text_upper = decision_text.upper()
if "COMPLETE" in decision_text_upper:
return RetrievalDecision.COMPLETE, None
elif "RETRIEVE" in decision_text_upper:
# Extract query from response
lines = decision_text.split('\n')
query_line = next((line for line in lines if "query" in line.lower()), None)
if query_line and ':' in query_line:
retrieval_query = query_line.split(':', 1)[1].strip()
else:
retrieval_query = query # Fallback to original query
return RetrievalDecision.RETRIEVE, retrieval_query
else:
return RetrievalDecision.CONTINUE_GENERATION, None
def _generate(self, query: str, context: str, previous: str) -> str:
"""Generate content with available context"""
response = self.generation_chain.invoke({
"query": query,
"context": context if context else "No additional context available",
"previous_generation": previous if previous else "Starting response"
})
return response
def demonstrate_retrieval_interleaving():
"""Demonstrate retrieval interleaving pattern"""
print("=" * 80)
print("RETRIEVAL INTERLEAVING PATTERN DEMONSTRATION")
print("=" * 80)
# Create agent with mock vector store
agent = RetrievalInterleavingAgent(max_retrievals=3)
# Example 1: Complex query requiring multiple retrievals
print("\n" + "=" * 80)
print("Example 1: Multi-Hop Question")
print("=" * 80)
query1 = "Explain how transformers relate to GPT models and why they're important for NLP"
print(f"\nQuery: {query1}")
print("\nExecuting with interleaved retrieval...")
print("-" * 60)
result1 = agent.generate_with_interleaved_retrieval(query1)
print(f"\nFinal Output:")
print(result1.final_output)
print(f"\n" + "-" * 60)
print(f"Execution Summary:")
print(f" Total retrievals: {result1.total_retrievals}")
print(f" Total generation steps: {len(result1.generation_steps)}")
print(f" Estimated tokens: {result1.total_tokens_estimated}")
print(f"\nRetrieval Steps:")
for i, step in enumerate(result1.retrieval_steps, 1):
print(f"\n Step {i} (timestamp {step.timestamp}):")
print(f" Query: {step.query}")
print(f" Retrieved {len(step.documents)} documents")
for j, doc in enumerate(step.documents, 1):
print(f" Doc {j}: {doc.page_content[:60]}...")
print(f"\nGeneration Steps:")
for i, step in enumerate(result1.generation_steps, 1):
used = "with retrieval" if step.used_retrieval else "without retrieval"
print(f" Step {i} (timestamp {step.timestamp}): Generated {used}")
print(f" Content: {step.content[:80]}...")
# Example 2: Simple query (fewer retrievals)
print("\n" + "=" * 80)
print("Example 2: Simple Query")
print("=" * 80)
query2 = "What is machine learning?"
print(f"\nQuery: {query2}")
result2 = agent.generate_with_interleaved_retrieval(query2)
print(f"\nFinal Output:")
print(result2.final_output)
print(f"\nExecution Summary:")
print(f" Total retrievals: {result2.total_retrievals}")
print(f" (Simpler query required fewer retrievals)")
# Example 3: Comparing with standard RAG
print("\n" + "=" * 80)
print("Example 3: Comparison - Interleaved vs Standard RAG")
print("=" * 80)
query3 = "Explain deep learning and how it relates to computer vision and NLP"
print(f"\nQuery: {query3}")
# Interleaved approach
print("\n1. Interleaved Retrieval Approach:")
result_interleaved = agent.generate_with_interleaved_retrieval(query3)
print(f" Retrievals: {result_interleaved.total_retrievals}")
print(f" Output: {result_interleaved.final_output[:150]}...")
# Standard RAG (single retrieval)
print("\n2. Standard RAG (single retrieval):")
docs = agent.vector_store.similarity_search(query3, k=3)
context = "\n\n".join([doc.page_content for doc in docs])
standard_prompt = ChatPromptTemplate.from_messages([
("system", "You are a helpful AI assistant. Use the provided context to answer the query."),
("user", "Context:\n{context}\n\nQuery: {query}\n\nAnswer:")
])
standard_chain = standard_prompt | agent.llm | StrOutputParser()
standard_result = standard_chain.invoke({"context": context, "query": query3})
print(f" Retrievals: 1")
print(f" Output: {standard_result[:150]}...")
print("\n Analysis:")
print(f" - Interleaved made {result_interleaved.total_retrievals} targeted retrievals")
print(f" - Standard RAG made 1 broad retrieval")
print(f" - Interleaved can gather more specific information as needed")
# Example 4: Retrieval pattern analysis
print("\n" + "=" * 80)
print("Example 4: Retrieval Pattern Analysis")
print("=" * 80)
queries = [
"What is NLP?",
"Explain the relationship between machine learning, deep learning, and transformers",
"How does reinforcement learning work?"
]
print("\nAnalyzing retrieval patterns for different query types:\n")
for query in queries:
result = agent.generate_with_interleaved_retrieval(query)
print(f"Query: {query}")
print(f" Retrievals: {result.total_retrievals}")
print(f" Generation steps: {len(result.generation_steps)}")
if result.retrieval_steps:
print(f" Retrieval queries:")
for step in result.retrieval_steps:
print(f" - {step.query}")
print()
# Summary
print("\n" + "=" * 80)
print("SUMMARY")
print("=" * 80)
print("""
The Retrieval Interleaving pattern enables:
✓ Dynamic, adaptive information gathering
✓ Multiple retrieval steps during generation
✓ More targeted and relevant retrievals
✓ Better handling of complex multi-hop queries
✓ Reduced initial retrieval overhead
✓ Context-aware retrieval decisions
✓ Improved accuracy for knowledge-intensive tasks
This pattern is valuable for:
- Complex research questions
- Multi-hop reasoning tasks
- Long-form content generation
- Dynamic information needs
- Iterative question answering
- Knowledge synthesis tasks
- Exploratory queries
Compared to standard RAG:
- Standard RAG: One retrieval upfront, simpler, faster
- Interleaved: Multiple targeted retrievals, more complex, better for complex queries
""")
if __name__ == "__main__":
demonstrate_retrieval_interleaving()
Agentic Design Pattern: Retrieval Interleaving
Pattern 169: Retrieval Interleaving – Line-by-Line Explanation
Overview
This document provides a detailed line-by-line explanation of the Retrieval Interleaving pattern implementation.
Pattern Description
Pattern 169: Retrieval Interleaving
Description: The Retrieval Interleaving pattern dynamically retrieves information during generation, interleaving retrieval steps with reasoning and generation steps. Unlike standard RAG which retrieves once upfront, this pattern retrieves multiple times as needed based on the generation process.
Components: 1. Generation Monitor: Tracks generation progress 2. Retrieval Trigger: Decides when to retrieve 3. Query Generator: Creates retrieval queries 4. Retriever: Fetches relevant information 5. Context Integrator: Merges retrieved content 6. Iteration Controller: Manages interleaving loops
Use Cases: – Complex multi-hop reasoning – Long-form content generation – Research-intensive tasks – Dynamic information needs – Iterative question answering – Knowledge synthesis
Benefits: – Adaptive information gathering – Better context relevance – Handles complex queries – Reduces initial retrieval overhead – More accurate results
Trade-offs: – Multiple retrieval calls – Increased latency – Complex orchestration – Higher costs – Requires good triggering logic
LangChain Implementation: Implements iterative retrieval with LangChain’s retrieval components. Uses LLM to decide when and what to retrieve during generation.
File Statistics
- Total Lines: 453
- Classes: 6
- Functions: 1
- Constants: 0
Module Documentation
"""
Pattern 169: Retrieval Interleaving
Description:
The Retrieval Interleaving pattern dynamically retrieves information during
generation, interleaving retrieval steps with reasoning and generation steps.
Unlike standard RAG which retrieves once upfront, this pattern retrieves
multiple times as needed based on the generation process.
Components:
1. Generation Monitor: Tracks generation progress
2. Retrieval Trigger: Decides when to retrieve
3. Query Generator: Creates retrieval queries
4. Retriever: Fetches relevant information
5. Context Integrator: Merges retrieved content
6. Iteration Controller: Manages interleaving loops
Use Cases:
- Complex multi-hop reasoning
- Long-form content generation
- Research-intensive tasks
- Dynamic information needs
- Iterative question answering
- Knowledge synthesis
Benefits:
- Adaptive information gathering
- Better context relevance
- Handles complex queries
- Reduces initial retrieval overhead
- More accurate results
Trade-offs:
- Multiple retrieval calls
- Increased latency
- Complex orchestration
- Higher costs
- Requires good triggering logic
LangChain Implementation:
Implements iterative retrieval with LangChain's retrieval components.
Uses LLM to decide when and what to retrieve during generation.
"""
Purpose: This module implements the Retrieval Interleaving 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.documents.Document
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
import typing.Tuple
Purpose: Core Python functionality:
– dataclasses.dataclass: Simplified class definitions with @dataclass
– dataclasses.field: Simplified class definitions with @dataclass
– enum.Enum: Enumeration support
– langchain_core.documents.Document: Regular expression operations
– 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)
– typing.Tuple: Type hints for better code documentation (List, Dict, Any, Optional)
LangChain Imports
from langchain_openai.ChatOpenAI
from langchain_openai.OpenAIEmbeddings
Purpose: LangChain framework components: – LLM Interface: Wrapper for OpenAI models – Chains: Composable building blocks for complex workflows
Other Imports
from dotenv.load_dotenv
Purpose: Additional dependencies for specialized functionality.
Classes
Class: RetrievalDecision
Lines 59-63
Inherits from: Enum
Purpose:
Decision on whether to retrieve
Class: RetrievalStep
Lines 67-72
Purpose:
Represents a retrieval step
Class: GenerationStep
Lines 76-80
Purpose:
Represents a generation step
Class: InterleavedResult
Lines 84-90
Purpose:
Result of interleaved retrieval-generation
Class: MockVectorStore
Lines 93-130
Purpose:
Mock vector store for demonstration
Methods:
__init__(self)
Lines 96-116
Purpose:
Initialize with sample documents
Parameters:
similarity_search(self, query, k)
Lines 118-130
Purpose:
Mock similarity search
Parameters:
– query: Input query or question
– k: K
Returns: <ast.Subscript object at 0x000001A06E0ABAD0>
Class: RetrievalInterleavingAgent
Lines 133-297
Purpose:
Agent that interleaves retrieval with generation
Methods:
__init__(self, vector_store, max_retrievals)
Lines 136-182
Purpose:
Initialize interleaving agent
Args:
vector_store: Vector store for retrieval
max_retrievals: Maximum number of retrieval steps
Parameters:
– vector_store: Vector store
– max_retrievals: Max retrievals
generate_with_interleaved_retrieval(self, query)
Lines 184-262
Purpose:
Generate response with interleaved retrieval
Parameters:
– query: Input query or question
Returns: InterleavedResult
_decide_next_step(self, query, current_generation, context)
Lines 264-288
Purpose:
Decide whether to retrieve, continue, or complete
Parameters:
– query: Input query or question
– current_generation: Current generation
– context: Context
Returns: <ast.Subscript object at 0x000001A06D60AC90>
_generate(self, query, context, previous)
Lines 290-297
Purpose:
Generate content with available context
Parameters:
– query: Input query or question
– context: Context
– previous: Previous
Returns: str
Functions
demonstrate_retrieval_interleaving()
Lines 300-448
Purpose:
Demonstrate retrieval interleaving pattern
Key Concepts Summary
Pattern Overview
The Retrieval Interleaving 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 168
- Pattern 170
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 Retrieval Interleaving 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: 169_retrieval_interleaving.py
Total Lines: 453
Pattern Category: AI Agent Design Pattern #169
Generated from actual implementation code
For educational and reference purposes