Who is this for?

Anyone curating before/after text examples in a spreadsheet and wanting a push-button path to a fine-tuned GPT model—without touching curl. Works with Google Sheets or Airtable.

What problem does it solve?

Manually downloading CSVs, converting to JSONL, uploading, and polling OpenAI is a slog.
This flow automates the whole loop: grab examples flagged Ready, build the JSONL file, start the fine-tune, then log the resulting model ID back to a registry sheet/table for reuse.

How it works

#NodePurpose1Schedule TriggerRuns weekly by default (change as needed).2aGet Examples from SheetPulls rows where Ready = TRUE from your Google Sheet. Uses the JSONL-Template Sheet as the expected column layout.2bGet Examples from Airtable (disabled)Alternate source for Airtable users.3Create JSONL File (Code)Converts each example to chat-format JSONL and splits into train.jsonl / val.jsonl (80/20).4Upload JSONLUploads the training file to OpenAI (purpose: fine-tune).5Begin Fine-TuneStarts a fine-tune job on gpt-4o (editable).6Wait → Check Job → IFPolls every minute until status = succeeded.7aWrite Model to SheetAppends the new model ID + meta to your Model Registry sheet.7bWrite Model to Airtable (disabled)Equivalent logging step for Airtable.

Setup steps

1. Import & connect credentials
   • Import the JSON flow into n8n.
   • Add your OpenAI API key.
   • Google Sheets: create an OAuth2 credential and link it to both Sheets nodes.
   • Airtable (optional): create a Personal Access Token and attach it to the Airtable nodes.
2. Copy the template sheet
   • Duplicate the JSONL-Template Sheet linked above into your own Drive.
   • Required columns (exact names):
     | systemPrompt | userPrompt | assistantResponse | Ready |
   • Tick Ready = TRUE for rows you want to include.
3. Create the registry sheet/table
   • Google Sheet (or Airtable table) named Model Registry with columns:
     Model ID, Training Examples, Epochs, Batch Size, Learning Rate, Finished At.
4. Tweak model & schedule
   • Change the base model in Begin Fine-Tune if desired.
   • Adjust the Schedule Trigger for daily / on-demand runs.
5. Test it
   • Mark a few examples Ready = TRUE.
   • Run the flow manually.
   • Check OpenAI for the new fine-tune job and confirm the model ID is logged in your registry.

Resources

• n8n Docs – <https://docs.n8n.io/>
• OpenAI Fine-Tuning – <https://platform.openai.com/docs/guides/fine-tuning>
• Google Sheets API – <https://developers.google.com/sheets/api>
• Airtable API – <https://airtable.com/api>

Extending the flow

• Webhook trigger – swap the schedule for a webhook to train on demand.
• Multi-source merge – enable both Sheets and Airtable nodes to combine datasets.
• Auto-deploy – save the new model name to an env-var or Secrets Manager for downstream generation workflows.