Structured Output with Union Events API

This repository contains a small CLI tool that demonstrates how to parse raw log lines into typed events using LangChain's structured output capabilities and Pydantic v2.

Features

• Two event models:
  • HttpOkEvent – successful HTTP request (status 200)
  • HttpErrorEvent – error response (4xx/5xx)
• Uses a discriminator field (kind) to route the parsed data into the correct model.
• Parses each log line with an LLM via LangChain and returns a validated Pydantic object.
• Prints a simple table of all parsed events.

Requirements

langchain-core>=1.0
langchain-openai
pydantic>=2
python-dotenv

Install them with:

pip install -r requirements.txt

Note

: The script uses the gpt-4o-mini model by default. Set the environment variable OPENAI_API_KEY to your key.

Usage

From a file

python main.py --file logs.txt

From stdin

echo -e "GET /api/v1/users 200 123ms\nPOST /api/v1/login 404 Not Found" | python main.py

The output will look like:

kind | path          | status | duration_ms | error_message
-----+---------------+--------+-------------+--------------
ok   | /api/v1/users | 200    | 123         |
error| /api/v1/login | 404    |             | Not Found

How it works

The script builds a LangChain chain that:

1. Prompts the LLM to parse a single log line into JSON.
2. Uses PydanticOutputParser to validate and convert the JSON into one of the two Pydantic models.
3. Collects all parsed events and prints them in a table.

The union type is handled automatically by LangChain's structured output parser thanks to the discriminator field.

License

MIT © 2026