Social Media Content Engine

Context

Aesthetic clinics needed consistent social media presence but producing quality content required a copywriter, designer, and community manager per clinic. Multiplied across N clinics, the operational cost was unsustainable.

Solution

An AI pipeline with a Social Media Copy Generator (Gemini), an Art Director agent that writes visual prompts aligned with each clinic's brand identity, fal.ai image generation with async queue polling, and a self-healing retry loop with Gemini Vision validation. Output lands as a ready-to-approve draft in Publer with Slack notification.

Results

• From hours to minutes — copy, design, and publishing in one autonomous flow
• Scales to N clinics with zero marginal cost
• Self-healing: failed images regenerate automatically (configurable max retries)
• Visual Operating System in Airtable — brand updates reflected immediately
• Human approval preserved: drafts in Publer, one click to publish

Architecture

Sub-workflow Trigger → Stage 1: Brand Context
  Airtable (Visual Operating System — brand guidelines per clinic)

→ Stage 2: Format Routing (Switch)
  Carousel → Copy Generator A → Slide Splitter A
  Stories  → Copy Generator B → Slide Splitter B

→ Stage 3: Copy Generation (per format)
  Gemini (Social Media Copy Generator)
  → Caption + slide structure

→ Stage 4: Art Direction & Image Generation
  Gemini (Art Director) → visual prompt per slide
  → fal.ai nano-banana-pro (async queue)
    → Poll status until COMPLETED
    → Download image binary

→ Stage 5: Visual Validation Loop
  Gemini Vision (analyze image)
  → Approved? → Publer media upload
  → Rejected? → retryCount < maxRetries? → regenerate
    → No? → skip slide

→ Stage 6: Draft Creation
  Publer API (POST /posts/schedule)
  → Poll job status until complete

→ Stage 7: Notification
  Slack (chat.postMessage) → Team notified