Skip to main content
When interacting with a language model, the raw output is always returned as a string. Depending on your application, you’ll need to parse, validate, and reformat this text before using it in your codebase.

​
Receiving the LLM Response

Typically, sending a prompt (and optional context) looks like this: 
# Send prompt to LLM and receive a raw string response
response = llm.generate(prompt, context=optional_context)
print(type(response))  # <class 'str'>

​
Formatting and Parsing LLM Output

Below are patterns to transform raw LLM responses into structured data.

​
Example: Calculating Date Differences

Suppose the LLM returns two dates in plain text: 
Start date: 2023-07-15
End date:   2023-08-01
You can extract and compute the difference with Python: 
from datetime import datetime

# Simulated LLM output
llm_output = "Start date: 2023-07-15\nEnd date:   2023-08-01"

# Split lines and extract date strings
lines = llm_output.splitlines()
start_str = lines[0].split(":")[1].strip()
end_str   = lines[1].split(":")[1].strip()

# Parse to datetime objects
start_date = datetime.strptime(start_str, "%Y-%m-%d")
end_date   = datetime.strptime(end_str,   "%Y-%m-%d")

# Compute and display difference
delta = end_date - start_date
print(f"Number of days between dates: {delta.days}")
Always strip() your strings to remove extra whitespace or escape characters before parsing.

​
Structuring Output in Desired Formats

As an alternative, ask the LLM to return JSON for easier parsing: 
{
  "start_date": "2023-07-15",
  "end_date":   "2023-08-01"
}
Then you can load and process it: 
import json
from datetime import datetime

json_str = '''
{"start_date": "2023-07-15", "end_date": "2023-08-01"}
'''
data = json.loads(json_str)
start_date = datetime.fromisoformat(data["start_date"])
end_date   = datetime.fromisoformat(data["end_date"])

print((end_date - start_date).days)
You can apply similar strategies with other formats:
FormatLibraryUse Case
JSONjsonFast and ubiquitous
YAMLPyYAMLHuman-readable configuration
XMLxml.etreeInteroperability with legacy
CSVcsvTabular data exchange
Choose the format that best fits your downstream requirements and existing data pipeline.

​
Cleaning and Validating the Output

A robust post-processing pipeline should:
  • Trim whitespace and remove unwanted escape characters
  • Validate date formats (e.g., using datetime.strptime)
  • Check numeric ranges and required fields
  • Handle missing or malformed values gracefully
Skipping validation can lead to runtime errors or corrupted data in production.

​
Next Steps: Model I/O Patterns

Explore best practices for serializing and deserializing data when calling language models. Learn how to:
  • Stream outputs incrementally
  • Implement retry logic and error handling
  • Cache or batch requests for performance

Watch Video