Benchmark LLM performance on legal documents with Google Sheets and OpenRouter

by Adam Janes ✓ · 860 views · created 2025-06-20 · Engineering, AI Summarization

What this workflow does

This AI workflow has 15 nodes and runs manually, when you click “Execute workflow” (plus 1 other trigger). Along the way it connects Google Sheets, Google Drive and OpenRouter Chat Model. It also calls webhook-processor-production-48f8.up.railway.app over HTTP.

How it runs

Trigger: Manually, when you click “Execute workflow” (plus 1 other trigger).

1. Get Tests (Google Sheets): uses Google Sheets
2. Keep Original Data (Edit Fields (Set)): sets or maps fields
3. Save Input/Output (Edit Fields (Set)): sets or maps fields
4. Is PDF? (If): checks a condition and branches into true / false
5. Merge: merges the data from its input branches
6. Google Drive: download
7. Limit: keeps the first few items
8. Extract from File: extracts data from a file (pdf)
9. Execute Subworkflow (HTTP Request): calls webhook-processor-production-48f8.up.railway.app (POST)
10. Basic LLM Chain: sends a prompt to an LLM, with model: OpenRouter Chat Model; output parser: Structured Output Parser

…plus 1 more step. See the full node list below.

Services it uses

Google SheetsGoogle DriveOpenRouter Chat Model

HTTP calls to: webhook-processor-production-48f8.up.railway.app

Before you upgrade to n8n 3.0

Nodes used (13)

• Manual Triggern8n-nodes-base.manualTrigger
• Webhookn8n-nodes-base.webhook
• Mergen8n-nodes-base.merge
• Google Sheetsn8n-nodes-base.googleSheets
• Ifn8n-nodes-base.if
• Structured Output Parser@n8n/n8n-nodes-langchain.outputParserStructured
• Basic LLM Chain@n8n/n8n-nodes-langchain.chainLlm
• Google Driven8n-nodes-base.googleDrive
• Extract from Filen8n-nodes-base.extractFromFile
• Edit Fields (Set)n8n-nodes-base.set
• HTTP Requestn8n-nodes-base.httpRequest
• OpenRouter Chat Model@n8n/n8n-nodes-langchain.lmChatOpenRouter
• Limitn8n-nodes-base.limit

From the author

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. This workflow extends the…

View the original on n8n.io → Check your own copy