Let's talk

Content automation

AI Social Video Pipeline

An end-to-end pipeline combining Claude, the Higgsfield MCP server and Supabase to produce UGC-style real estate videos for 7 social platforms.

Role
AI and backend engineering
Platforms
Pipeline · Dashboard
Stack
Claude · Higgsfield MCP · Supabase
  • Claude
  • MCP
  • Prompt versioning
  • Automation
Video pipeline run with renders for each platform and quality gates

Product screens with sample data. Names and figures inside the screens are illustrative.

01

The problem

Generated video fails quietly: silent audio, broken on-screen text. Fixing it one video at a time doesn't scale.

7
Social platforms
E2E
Automated
Versioned
Brand prompts
02

What I built

01

Prompts as code

Versioned brand system prompts and a structured prompt framework fixed the silent-video and text-rendering failures, and every render traces back to its inputs.

Run detail with platform renders and quality gates
03

Key decisions

  1. Version the prompt like a schema

    When output breaks, you can see which prompt version changed and roll it back.

Next project · 10Requirements IntelligenceAI pipeline