Back to home

Case Study — Ongoing

Running 3 Medical Aesthetics
Brands at Scale

同時經營 3 個醫美品牌的自動化作業系統

Industry

Medical Aesthetics

醫美產業

Scope

3 brands · 60+ pieces/mo

三品牌平行

Duration

2024 → Present

持續進行中

Role

System Architect + Operator

架構 + 實作

01

The Challenge

面對的挑戰

Three separate medical aesthetics brands. Each with its own visual identity, content calendar, audience, and compliance requirements. One operator. No agency budget. The question was never “can we post more?” — it was “can we build a system that posts correctly, consistently, and legally, without burning someone out?”

某醫美集團需要同時代管三個品牌,而台灣醫美廣告法規嚴格——傳統代操要不是量跟不上,就是踩到紅線。

三個獨立的醫美品牌,各自有視覺識別、內容行事曆、受眾與合規要求,只有一個操作者,沒有代理商預算。問題從來不是「能不能多發文」——而是「能不能建立一套系統,讓每篇貼文都正確、一致、合法,同時不讓人精疲力竭?」

02

The Architecture

系統架構

Five sequential stages, each automated to the degree possible. Human judgment is injected at brief-creation and final approval only — everything else runs on its own.

Step 1

Slides Design

Templates built in Google Slides. Brand palette, typography, and layout rules locked in. Content filled per brief.

Step 2

Compliance Scan

Claude Agents scan all copy against a 32-term medical-claim blacklist before anything goes near a scheduler.

Step 3

Auto-Publish

Cloudflare Worker handles Instagram + Facebook scheduled publishing via Meta Graph API. No manual posting.

Step 4

Insights Collection

Cron job pulls per-post metrics (reach, impressions, engagement) into a D1 SQLite database for reporting.

Step 5

Monthly Report

Auto-generated Google Slides report aggregates 30 days of data, highlights top posts, and suggests next-month pivots.

03

The Numbers

量化成果

60+

Pieces per month

Across 3 brands

3

Brands managed

Simultaneously

32

Compliance terms

In blacklist

0

Manual publish steps

After setup

04

Technical Deep Dive

技術細節

Every component was chosen for reliability at the edge, not convenience on a laptop. The goal was a system that runs at 2am without supervision.

Cloudflare Worker

Edge-deployed publish engine. Handles Meta API auth, retry logic, and scheduled post dispatch without a server.

Compliance Scanner

Claude Agents evaluate copy against a curated 32-term medical-claim blacklist. Blocks anything that violates Taiwan health regulations.

Video Generation

Veo 3.1 + local ffmpeg pipeline. Generates 15s reels, applies brand overlays, and exports to Cloudflare R2.

Insights API

Per-metric try/catch wrapper around Meta Graph API. Stores reach, impressions, and engagement in Cloudflare D1.

Cloudflare R2

Zero-egress object storage for all brand assets — images, videos, brand kits — accessible across all 3 brand pipelines.

Claude Agents

Multi-agent orchestration for compliance checking, caption rewriting, and insight summaries. Sonnet for drafts, Haiku for bulk.

05

What This Proves

這套系統證明了什麼

  • A single engineer can operate 3 active brands at production quality — if the system is designed right.

  • Compliance automation is not optional in regulated industries. It must be upstream of publishing, not a manual review step.

  • Edge-deployed automation (Cloudflare Workers + D1) eliminates hosting costs and cold-start latency for social pipelines.

  • Structured data collection from day one turns monthly reporting from a 4-hour task into a 10-minute review.

Need a system like this?

需要這樣的系統?

Get in touch →