Personal Brand Engine
Automated content generation and management through a sub-domain dashboard.
Skills
Claude Code · Claude Agent SDK · Platform Configuration
Year
2026
Status
Active
Overview
Personal Brand Engine is a 21-agent system running on a Hetzner VPS that writes, schedules, and learns from my content across LinkedIn, X, YouTube, and Instagram. Every weekday it runs a voice or text interview with me, then fans the brief out to per-platform Strategist, Writer, and Designer chains.
The problem
A founder who ships across the stack also has to be visible, and visibility is its own job. I needed a system that could compound my output across four platforms without burning the cognitive load I needed for actual product work. Off-the-shelf social tools were either generic schedulers or shallow AI wrappers.
Process
- 01
Built the interview agent in Claude Opus 4.6 — a daily voice or text session that captures the brief
- 02
Stood up a Hetzner CPX21 VPS with n8n for orchestration, SQLite for state, PM2 for the dashboard, and Caddy for TLS
- 03
Wrote per-platform Strategist, Writer, and Designer chains for LinkedIn, X, YouTube, and Instagram
- 04
Wired the learning loop: every edit, post, and performance metric flows back into SQLite, scored by confidence and decayed over time
- 05
Built the Next.js review dashboard at brand.copytier.com, single-user behind a Google OAuth allowlist
- 06
Rebuilt the X engine as engagement-first after the first version posted 18-22 tweets a week with zero results
Solution
A 21-agent pipeline with file-based handoffs — markdown in, markdown out — coordinated by n8n. Interview agent at the top, per-platform chains in the middle, Editor-learner and Performance-learner closing the loop. The dashboard surfaces everything queued and everything posted, so I approve before anything ships.
Results
The hardest lesson came from the X engine. The first version pumped out 18-22 tweets a week and earned nothing. I tore it down and rebuilt it as an engagement-first system with a weekly ops brief telling me whose timelines to live in. The metrics were telling me something my prompt design wasn't, and I listened.