langchain-orchestration

What this skill does

# LangChain Orchestration Skill

Complete guide for building production-grade LLM applications with LangChain, covering chains, agents, memory, RAG patterns, and advanced orchestration techniques.

## Table of Contents

1. [Core Concepts](#core-concepts)
2. [Chains](#chains)
3. [Agents](#agents)
4. [Memory Systems](#memory-systems)
5. [RAG Patterns](#rag-patterns)
6. [LLM Integrations](#llm-integrations)
7. [Callbacks & Monitoring](#callbacks--monitoring)
8. [Retrieval Strategies](#retrieval-strategies)
9. [Streaming](#streaming)
10. [Error Handling](#error-handling)
11. [Production Best Practices](#production-best-practices)

## Core Concepts

### LangChain Expression Language (LCEL)

LCEL is the declarative way to compose chains in LangChain, enabling streaming, async, and parallel execution.

```python
from langchain_core.runnables import RunnablePassthrough
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_openai import ChatOpenAI

# Basic LCEL chain
prompt = ChatPromptTemplate.from_template("Tell me about {topic}")
llm = ChatOpenAI(model="gpt-4o-mini", temperature=0)
output_parser = StrOutputParser()

chain = prompt | llm | output_parser
result = chain.invoke({"topic": "quantum computing"})
```

### Runnable Interface

Every component in LangChain implements the Runnable interface with standard methods:

```python
from langchain_core.runnables import RunnablePassthrough

# Key methods: invoke, stream, batch, ainvoke, astream, abatch
chain = prompt | llm | output_parser

# Synchronous invoke
result = chain.invoke({"topic": "AI"})

# Streaming
for chunk in chain.stream({"topic": "AI"}):
    print(chunk, end="", flush=True)

# Batch processing
results = chain.batch([{"topic": "AI"}, {"topic": "ML"}])

# Async variants
result = await chain.ainvoke({"topic": "AI"})
```

### RunnablePassthrough

Pass inputs directly through or apply transformations:

```python
from langchain_core.runnables import RunnablePassthrough

# Pass through unchanged
chain = RunnablePassthrough() | llm | output_parser

# With transformation
def add_context(x):
    return {"text": x["input"], "context": "important"}

chain = RunnablePassthrough.assign(processed=add_context) | llm
```

## Chains

### Sequential Chains

Process data through multiple steps sequentially.

```python
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(temperature=0)

# Step 1: Generate ideas
idea_prompt = ChatPromptTemplate.from_template(
    "Generate 3 creative ideas for: {topic}"
)
idea_chain = idea_prompt | llm | StrOutputParser()

# Step 2: Evaluate ideas
eval_prompt = ChatPromptTemplate.from_template(
    "Evaluate these ideas and pick the best one:\n{ideas}"
)
eval_chain = eval_prompt | llm | StrOutputParser()

# Combine into sequential chain
sequential_chain = (
    {"ideas": idea_chain}
    | RunnablePassthrough.assign(evaluation=eval_chain)
)

result = sequential_chain.invoke({"topic": "mobile app"})
```

### Map-Reduce Chains

Process multiple inputs in parallel and combine results.

```python
from langchain_core.runnables import RunnableParallel
from langchain_core.prompts import ChatPromptTemplate

# Define parallel processing
summary_prompt = ChatPromptTemplate.from_template(
    "Summarize this text in one sentence: {text}"
)
keywords_prompt = ChatPromptTemplate.from_template(
    "Extract 3 keywords from: {text}"
)
sentiment_prompt = ChatPromptTemplate.from_template(
    "Analyze sentiment (positive/negative/neutral): {text}"
)

# Map: Process in parallel
map_chain = RunnableParallel(
    summary=summary_prompt | llm | StrOutputParser(),
    keywords=keywords_prompt | llm | StrOutputParser(),
    sentiment=sentiment_prompt | llm | StrOutputParser()
)

# Reduce: Combine results
reduce_prompt = ChatPromptTemplate.from_template(
    """Combine the analysis:
    Summary: {summary}
    Keywords: {keywords}
    Sentiment: {sentiment}

    Provide a comprehensive report:"""
)

map_reduce_chain = map_chain | reduce_prompt | llm | StrOutputParser()

result = map_reduce_chain.invoke({
    "text": "LangChain is an amazing framework for building LLM applications."
})
```

### Router Chains

Route inputs to different chains based on conditions.

```python
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser

# Define specialized chains
technical_prompt = ChatPromptTemplate.from_template(
    "Provide a technical explanation of: {query}"
)
simple_prompt = ChatPromptTemplate.from_template(
    "Explain in simple terms: {query}"
)

technical_chain = technical_prompt | llm | StrOutputParser()
simple_chain = simple_prompt | llm | StrOutputParser()

# Router function
def route_query(input_dict):
    query = input_dict["query"]
    complexity = input_dict.get("complexity", "simple")

    if complexity == "technical":
        return technical_chain
    return simple_chain

# Create router chain
from langchain_core.runnables import RunnableLambda

router_chain = RunnableLambda(route_query)

# Use the router
result = router_chain.invoke({
    "query": "quantum entanglement",
    "complexity": "technical"
})
```

### Conditional Chains

Execute chains based on conditions.

```python
from langchain_core.runnables import RunnableBranch

# Define condition-based routing
classification_prompt = ChatPromptTemplate.from_template(
    "Classify this as 'question', 'statement', or 'command': {text}"
)

question_handler = ChatPromptTemplate.from_template(
    "Answer this question: {text}"
) | llm | StrOutputParser()

statement_handler = ChatPromptTemplate.from_template(
    "Acknowledge this statement: {text}"
) | llm | StrOutputParser()

command_handler = ChatPromptTemplate.from_template(
    "Execute this command: {text}"
) | llm | StrOutputParser()

# Create conditional branch
branch = RunnableBranch(
    (lambda x: "question" in x["type"].lower(), question_handler),
    (lambda x: "statement" in x["type"].lower(), statement_handler),
    command_handler  # default
)

# Full chain with classification
full_chain = (
    {"text": RunnablePassthrough(), "type": classification_prompt | llm | StrOutputParser()}
    | branch
)
```

### LLMChain (Legacy)

Traditional chain format still supported:

```python
from langchain.chains import LLMChain
from langchain_core.prompts import PromptTemplate

prompt = PromptTemplate(
    input_variables=["product"],
    template="What is a good name for a company that makes {product}?"
)

chain = LLMChain(llm=llm, prompt=prompt)
result = chain.run(product="eco-friendly water bottles")
```

### Stuff Documents Chain

Combine documents into a single context:

```python
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain_core.documents import Document

prompt = ChatPromptTemplate.from_template(
    """Answer based on the following context:

<context>
{context}
</context>

Question: {input}"""
)

document_chain = create_stuff_documents_chain(llm, prompt)

docs = [
    Document(page_content="LangChain supports multiple LLM providers."),
    Document(page_content="Chains can be composed using LCEL.")
]

result = document_chain.invoke({
    "input": "What does LangChain support?",
    "context": docs
})
```

## Agents

### ReAct Agents

Reasoning and Acting agents that use tools iteratively.

```python
from langchain.agents import create_react_agent, AgentExecutor
from langchain_core.tools import Tool
from langchain import hub

# Define tools
def search_tool(query: str) -> str:
    """Search for information"""
    return f"Search results for: {query}"

def calculator_tool(expression: str) -> str:
    """Calculate mathematical expressions"""
    try:
        return str(eval(expression))
    except:
        return "Invalid expression"

tools = [
    Tool(
        name="Search",
        func=search_tool,