How to Parse Structured Output in Langchain: Simple Guide

In Langchain, you parse structured output by using OutputParser classes like StructuredOutputParser or RegexParser to convert raw model responses into usable Python objects. These parsers help extract data cleanly by defining expected formats or patterns.

Syntax

To parse structured output in Langchain, you typically use an OutputParser instance. The main parts are:

• OutputParser: Base class for parsing outputs.
• parse() method: Takes raw string output and returns structured data.
• Specific parsers like StructuredOutputParser or RegexParser define how to extract data.

python

from langchain.output_parsers import StructuredOutputParser

# Create a parser with a defined schema
parser = StructuredOutputParser.from_response_schemas([
    {"name": "name", "description": "Person's name"},
    {"name": "age", "description": "Person's age"}
])

# Use parser.parse() to convert raw output
raw_output = '{"name": "Alice", "age": 30}'
parsed = parser.parse(raw_output)
print(parsed)

Output

{'name': 'Alice', 'age': 30}

Example

This example shows how to use StructuredOutputParser to parse a JSON-like string output from a language model into a Python dictionary with specific fields.

python

from langchain.output_parsers import StructuredOutputParser

# Define the expected output schema
schemas = [
    {"name": "city", "description": "Name of the city"},
    {"name": "temperature", "description": "Current temperature in Celsius"}
]

# Create the parser
parser = StructuredOutputParser.from_response_schemas(schemas)

# Simulated raw output from a language model
raw_output = '{"city": "Paris", "temperature": 18}'

# Parse the output
parsed_output = parser.parse(raw_output)
print(parsed_output)

Output

{'city': 'Paris', 'temperature': 18}

Common Pitfalls

Common mistakes when parsing structured output in Langchain include:

• Not matching the output format exactly, causing parse errors.
• Using raw string output without cleaning or validating it first.
• Ignoring exceptions raised by parse() when output is malformed.
• Confusing the parser's expected schema with the actual output format.

Always ensure your language model outputs match the parser's expected structure.

python

from langchain.output_parsers import StructuredOutputParser

schemas = [{"name": "name", "description": "Person's name"}]
parser = StructuredOutputParser.from_response_schemas(schemas)

# Wrong output format (missing quotes around name)
raw_output_wrong = '{name: Alice}'

try:
    parser.parse(raw_output_wrong)
except Exception as e:
    print(f"Error parsing output: {e}")

# Correct output format
raw_output_right = '{"name": "Alice"}'
parsed = parser.parse(raw_output_right)
print(parsed)

Output

Error parsing output: Expecting property name enclosed in double quotes: line 1 column 2 (char 1)
{'name': 'Alice'}

Quick Reference

Tips for parsing structured output in Langchain:

• Use StructuredOutputParser.from_response_schemas() to define expected fields.
• Always validate or clean raw output before parsing.
• Handle exceptions from parse() to catch malformed outputs.
• Test your parser with sample outputs to ensure compatibility.

Key Takeaways

Use Langchain's OutputParser classes to convert raw model output into structured data.

Define clear schemas matching expected output fields for reliable parsing.

Always handle parsing errors to avoid crashes from unexpected output formats.

Test parsers with sample outputs to ensure they work as intended.