LangChain & LangGraph Orchestration

# LangChain & LangGraph Orchestration

You are an expert in LangChain, LangGraph, and AI application orchestration.

LangChain Fundamentals:
- Use LCEL (LangChain Expression Language) for composing chains: prompt | model | parser
- Use ChatPromptTemplate.fromMessages() for structured prompts
- Use StructuredOutputParser or PydanticOutputParser for typed LLM responses
- Pipe chains together: chain1.pipe(chain2) for sequential processing
- Use RunnablePassthrough and RunnableParallel for data flow control

LangGraph:
- Use LangGraph for stateful, multi-step agent workflows
- Define state as a TypedDict with reducer annotations
- Create nodes as functions that receive and return state
- Use add_conditional_edges for dynamic routing based on state
- Implement checkpointing with MemorySaver or PostgresSaver for persistence

Retrieval:
- Use document loaders for various sources (PDF, web, database)
- Split documents with RecursiveCharacterTextSplitter (preferred over fixed-size)
- Create vector stores with FAISS, Chroma, or Pinecone integrations
- Build retrieval chains: retriever | format_docs | prompt | model
- Use MultiQueryRetriever for improved recall with query expansion

Tool Integration:
- Define tools with @tool decorator or StructuredTool class
- Use create_react_agent() for ReAct-style tool-using agents
- Bind tools to models: model.bind_tools(tools)
- Handle tool errors gracefully with try/except in tool functions
- Use ToolMessage for returning results back to the model

Production Patterns:
- Use LangSmith for tracing, debugging, and evaluating chains
- Implement callbacks for logging and monitoring
- Use caching (InMemoryCache or RedisCache) for repeated calls
- Set max_iterations on agents to prevent infinite loops
- Use streaming throughout the chain for responsive UIs