LangChain Core Workflow A: Chains & Prompts

Overview

Build production-ready chains using LangChain Expression Language (LCEL) with prompt templates, output parsers, and composition patterns.

Prerequisites

Completed langchain-install-auth setup
Understanding of prompt engineering basics
Familiarity with Python type hints

Instructions

Step 1: Create Prompt Templates

from langchain_core.prompts import (
    ChatPromptTemplate,
    SystemMessagePromptTemplate,
    HumanMessagePromptTemplate,
    MessagesPlaceholder
)

# Simple template
simple_prompt = ChatPromptTemplate.from_template(
    "Translate '{text}' to {language}"
)

# Chat-style template
chat_prompt = ChatPromptTemplate.from_messages([
    SystemMessagePromptTemplate.from_template(
        "You are a {role}. Respond in {style} style."
    ),
    MessagesPlaceholder(variable_name="history", optional=True),
    HumanMessagePromptTemplate.from_template("{input}")
])

Step 2: Build LCEL Chains

from langchain_openai import ChatOpenAI
from langchain_core.output_parsers import StrOutputParser, JsonOutputParser

llm = ChatOpenAI(model="gpt-4o-mini")

# Basic chain: prompt -> llm -> parser
basic_chain = simple_prompt | llm | StrOutputParser()

# Invoke the chain
result = basic_chain.invoke({
    "text": "Hello, world!",
    "language": "Spanish"
})
print(result)  # "Hola, mundo!"

Step 3: Chain Composition

from langchain_core.runnables import RunnablePassthrough, RunnableParallel

# Sequential chain
chain1 = prompt1 | llm | StrOutputParser()
chain2 = prompt2 | llm | StrOutputParser()

sequential = chain1 | (lambda x: {"summary": x}) | chain2

# Parallel execution
parallel = RunnableParallel(
    summary=prompt1 | llm | StrOutputParser(),
    keywords=prompt2 | llm | StrOutputParser(),
    sentiment=prompt3 | llm | StrOutputParser()
)

results = parallel.invoke({"text": "Your input text"})
# Returns: {"summary": "...", "keywords": "...", "sentiment": "..."}

Step 4: Branching Logic

from langchain_core.runnables import RunnableBranch

# Conditional branching
branch = RunnableBranch(
    (lambda x: x["type"] == "question", question_chain),
    (lambda x: x["type"] == "command", command_chain),
    default_chain  # Fallback
)

result = branch.invoke({"type": "question", "input": "What is AI?"})

Output

Reusable prompt templates with variable substitution
Type-safe LCEL chains with clear data flow
Composable chain patterns (sequential, parallel, branching)
Consistent output parsing

Examples

Multi-Step Processing Chain

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

llm = ChatOpenAI(model="gpt-4o-mini")

# Step 1: Extract key points
extract_prompt = ChatPromptTemplate.from_template(
    "Extract 3 key points from: {text}"
)

# Step 2: Summarize
summarize_prompt = ChatPromptTemplate.from_template(
    "Create a one-sentence summary from these points: {points}"
)

# Compose the chain
chain = (
    {"points": extract_prompt | llm | StrOutputParser()}
    | summarize_prompt
    | llm
    | StrOutputParser()
)

summary = chain.invoke({"text": "Long article text here..."})

With Context Injection

from langchain_core.runnables import RunnablePassthrough

def get_context(input_dict):
    """Fetch relevant context from database."""
    return f"Context for: {input_dict['query']}"

chain = (
    RunnablePassthrough.assign(context=get_context)
    | prompt
    | llm
    | StrOutputParser()
)

result = chain.invoke({"query": "user question"})

Error Handling

| Error | Cause | Solution | |-------|-------|----------| | Missing Variable | Template variable not provided | Check input dict keys match template | | Type Error | Wrong input type | Ensure inputs match expected schema | | Parse Error | Output doesn't match parser | Use more specific prompts or fallback |

Resources

Next Steps

Proceed to langchain-core-workflow-b for agents and tools workflow.