Marketing OS

The backend engine that turns a listing's raw photos and clips into a scheduled stream of branded, AI-produced video, with no editor in the loop.

Overview

Marketing OS (MOS) is Blonde Waterfall's automation engine for real-estate marketing. An agent uploads a listing once. MOS plans a content calendar, produces each video, and delivers it on schedule.

Manual editing was slow, expensive, and the bottleneck between agents and the steady stream of content they need.

Core workflow

• Listing creation and asset upload to S3.
• AI-generated content calendar, sent to the agent for approval.
• An SQS-driven production pipeline: image editing, frame expansion, AI video generation, shot reordering, then rendering.
• Final renders in After Effects through nexrender on a Windows render fleet, then delivery.
• Observability and typed retries at every step.

Who uses it

• Agents: upload listings, approve calendars, receive videos.
• Admins: credits, accounts, and operations.
• Editors: author and test templates in a separate editor app.
• An AI assistant that helps with planning and approvals.

Design choices

• One queue per pipeline step, so each stage scales and fails on its own.
• A single Docker image runs every worker type.
• Content-hash caching means an identical step never runs twice.
• A 48-hour production buffer keeps scheduled posts safe from late failures.
• Typed errors and stateless auth throughout.

Stack

• RUNTIME: Node.js + TypeScript
• QUEUES / STORAGE: AWS SQS, S3
• DATA / AUTH: Firestore, Firebase Auth
• GENERATIVE: Nano Banana (images), Veo (video)
• RENDER: After Effects + nexrender