LCEL Demo 1

In this guide, you’ll create a simple “Hello World” LCEL chain that connects a prompt template, a language model (LLM), and an output parser. You’ll then invoke the chain, inspect its input/output schemas, and explore individual component schemas.

Table of Contents

1. Prerequisites
2. Setup
3. Build the Chain
4. Invoke the Chain
5. Inspect Input Schema
6. Inspect Output Schema
7. Explore Component Schemas
8. Next Steps
9. References

1. Prerequisites

• Python 3.8+
• langchain-core and langchain-openai installed

pip install langchain-core langchain-openai

2. Setup

Import the necessary classes and define your prompt template, LLM, and output parser:

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

prompt = ChatPromptTemplate.from_template(
    """You are a helpful assistant.
    Answer the following question: {question}"""
)
llm = ChatOpenAI()
output_parser = StrOutputParser()

3. Build the Chain

Chain the components together using the pipe operator (|):

chain = prompt | llm | output_parser

4. Invoke the Chain

Provide the question input and call the chain:

response = chain.invoke({"question": "Tell me about The Godfather movie"})
print(response)

Example Output:

"The Godfather" is a classic crime film directed by Francis Ford Coppola and released in 1972. ...

5. Inspect Input Schema

Every LCEL chain exposes its input schema. To view it:

print(chain.input_schema.schema())

Sample Output:

{
  "title": "PromptInput",
  "type": "object",
  "properties": {
    "question": {
      "title": "Question",
      "type": "string"
    }
  }
}

6. Inspect Output Schema

Similarly, the chain’s output schema shows the format of the result:

print(chain.output_schema.schema())

Sample Output:

{
  "title": "StrOutputParserOutput",
  "type": "string"
}

7. Explore Component Schemas

You can inspect schemas for each component in the chain.

7.1 LLM Input Schema

print(llm.input_schema.schema())

Excerpt:

{
  "description": "Message from an AI.",
  "type": "object",
  "required": ["content"]
}

7.2 LLM Output Schema

print(llm.output_schema.schema())

Excerpt:

{
  "title": "ChatOpenAIOutput",
  "anyOf": [
    { "$ref": "#/definitions/AIMessage" },
    { "$ref": "#/definitions/HumanMessage" }
  ],
  "definitions": {
    "AIMessage": { "properties": { "content": { ... } }, "required": ["content"] },
    "HumanMessage": { "properties": { "content": { ... } } }
  }
}

8. Next Steps

• Integrate custom prompt templates and parsers
• Combine multiple LLMs or logic in a single chain
• Use the pass-through element to include custom Python code

References

• LangChain Documentation
• OpenAI API Reference