⚒️ Simple eval for legal benchmarking
⚡ 755 views · ⚒️ Engineering
Description
This workflow demonstrates a simple way to run evals on a set of test cases stored in a Google Sheet.
The example we are using comes from an info extraction task dataset, where we tested 6 different LLMs on 18 different test cases.
You can see our sample data in this spreadsheet here to get started.
Once you have this working for our dataset, you can plug in your own test cases matching different LLMs to see how it works with your own data.
How it works:
- It loads test cases from Google Sheets.
- For each row in our Google Sheet, it grabs the source document, converting it to text.
- Our “LLM judge” passes the input/output of each LLM to GPT-4.1 to evaluate each test case (Pass/Fail + Reason).
- It logs the outcome to a Google Sheet.
- A 0.5s pause between each request gets around OpenAI’s API rate limits.
Set up steps:
- Add your credentials for Google Sheets, Google Drive, and OpenRouter.
- Make a copy of the original data spreadsheet so that you can edit it yourself. You will need to plug your version in the Update Results node to see the spreadsheet update on each run of the loop.
🔗 Nodes Used
Google Sheets, Google Drive, Basic LLM Chain, Structured Output Parser, Extract from File, OpenRouter Chat Model
📥 Import
Download workflow.json and import into n8n:
Workflow menu → Import from File