A Software Factory on Kubernetes: Gemini Builds, jev Decides (Genkit + senro in Go)

Description and use

A Kubernetes software factory uses Gemini to generate code, jev to make decisions, and senro to orchestrate pipelines, automatically producing vetted Go modules from feature requests.

Problem

Automating end‑to‑end Go module creation from a feature request while ensuring quality and human oversight.

Outcome

The factory produces a fully tested, reviewed Go module with minimal human intervention, reducing manual review effort and ensuring consistent quality.

Workflow

• Receive feature request as request.md
• Design loop: Gemini generates design.json and DESIGN.md, jev reviews and loops until ready, revise, or reject
• Plan: Gemini produces plan.json and PLAN.md
• Implement: Gemini writes module files
• Vet: run go vet, capture vet.log and exit code
• Test: run go test, capture test.log and exit code
• Review: jev decides ship, rework, or human based on logs
• Rework: if needed, Gemini fixes code, repeat vet, test, review
• Finish: if ship, create PR.md and publish module to out/<request>/src; if human, create BRIEF.md and notify

Evidence

Everything below ran against the real Gemini API (gemini-3.8-flash) and the real TypeSafe API (jev-latest)

Source

Gemini writes every one of those. jev decides, at every gate, whether the work moves on, goes back, or stops for a human.

Source

The factory is three senro pipelines and a small state machine over them

Source