Before you press play
What you’ll find in the video
- Connect Jev in n8n using an HTTP request node via OpenRouter rather than standard chat completion nodes, since Jev exclusively outputs decision probabilities.
- Route classifications based on returned probability scores, routing lower-confidence evaluations (such as under 0.70) to human review.
- Jev only classifies among pre-defined options and cannot generate drafts, requiring separate templates or LLMs for automated replies.
Worth knowing
Gemini-assisted video/transcript review. The Gmail workflow is a demonstration, not a measured accuracy study. The example’s 0.70 threshold requires validation on real mail, and provider access and request formats can change.