Multi-agent on purpose
One mega-prompt lies. Separating generation from validation meant the “nice sounding” agent couldn’t bypass JSON or cultural checks. That’s the difference between a demo and something Novin could run.
Melipayamak · Novin Holdings · 2025
At Melipayamak / Novin I built an intelligent SMS template suggester. The point wasn’t “AI that writes texts” — it was stopping empty filter states from becoming user-facing 404s, while keeping tone culturally sane for Iranian SMS.
My role Automation architect — n8n multi-agent design, WP API integration, validation gates.
Generated templates were culturally flat, bad JSON leaked downstream, and empty filters 404’d the frontend. Marketing wanted speed; product needed guardrails.
n8n as the brain: WordPress backend hits a private API to trigger automation in real time. One agent family generates SMS samples from prompts + user filters; a second validates JSON integrity, relevance, and cultural sensitivity before anything ships.
Zero empty-filter 404s on that path, and SMS copy that doesn’t sound like English run through a blender.
One mega-prompt lies. Separating generation from validation meant the “nice sounding” agent couldn’t bypass JSON or cultural checks. That’s the difference between a demo and something Novin could run.
Ops already lived in WP. The private API bridge meant product UI could request suggestions without teaching everyone n8n.
Private API triggers so suggestions happen in the flow, not in a batch dump.
Tone/structure first; JSON + cultural checks second.
Frontend never eats a hollow payload that becomes a 404.