CLAYMR Pipeline
An onboarding and content pipeline for a Berlin restaurant-marketing agency. New client launched in 30 minutes, not a full day.
Every new restaurant onboarded to the agency required a day of manual content prep — scraping their Instagram, lifting menu PDFs, learning their tone, drafting first posts. The pipeline collapses that into a single webhook trigger and a 30-minute end-to-end run.
Pipeline
Six stages, one webhook.
A signed webhook fires when a new restaurant is added in the agency CRM. Payload includes IG handle, GMB URL and menu PDF link.
IG posts, Google reviews, menu PDF and any uploaded brand documents are pulled into a per-restaurant raw-content bucket.
Claude Vision lifts menu items, prices, photos and brand-language signals from PDFs and images into structured records.
Tone, slang, do-not-say list and preferred CTAs distilled into a typed Postgres record that conditions every future generation.
Menu copy, IG captions and GMB updates produced from the voice profile. Every output cost-tagged and budget-capped per restaurant.
Generated assets arrive in Notion as structured pages with approve/reject controls. Approved items push directly to scheduled IG / GMB / web.
/ Six stages, one pipeline
From signup to first published post in thirty minutes.
01 · Approach
The voice profile is the asset.
The cheap trick is to ask Claude to "write like a restaurant in Berlin." The real lever is building a voice profile once — extracted from every existing piece of content the restaurant has — then conditioning every future generation on it.
Profile lives in Postgres. Each restaurant becomes a tenant. Updates cascade through future runs but don't invalidate prior outputs.
Stack
Cost-tagged at every stage.
Reviewer-first handoff
The reviewer never has to leave Notion.
Every generated asset arrives as a structured page with restaurant context, voice-profile excerpts, and approve/reject controls. Approved items push directly to scheduled IG / GMB / web copy. Failed stages auto-retry with backoff before escalating.
From one onboarding per day to one per hour.
