Genkit Flows on Kubernetes, Routed by a Decision Model: senro + jev in Go

Description and use

A workflow that runs Genkit flows on Kubernetes with senro, using jev for routing and safety decisions and Gemini for writing replies, achieving fast, typed decisions and isolated pod execution.

Problem

Using a chat model alone to route support tickets and verify replies is slow, uncertain, and prone to errors.

Outcome

The pipeline achieves millisecond routing decisions, typed safety verification, and isolated pod execution with retries and caching, improving reliability and reducing latency compared to a pure chat‑model approach.

Workflow

• Seed raw email into run workspace
• Ingest email with Gemini to produce ticket.json
• Triage with jev to decide department, urgency, frustration, human need
• Collect ticket.json and decision.json
• Draft reply with Gemini per department playbook
• Verify draft with jev against playbook
• Send reply if verification passes
• Escalate to human with brief if jev indicates
• Skip steps based on earlier decisions

Evidence

Gemini writes. jev decides.

Source

A decision model is built for exactly that call.

Source

senro’s Kubernetes executor never reads KUBECONFIG, ~/.kube/config or your current context.

Source