WorkContent and ops fleet
Content and ops fleet
An automated content and operations system that drafts, schedules, replies and reports for 11 brands, with approval gates.
- Role
- Architecture and engineering
- Status
- Running daily
- Stack
- Node.js, Python, launchd, MCP, Codex, Local models
The problem
Running several small products alone leaves no time for marketing and operations. I wanted the routine work done by scheduled agents, with a person approving anything public or irreversible.
What I did
- Built Spellcast, a self-hosted multi-brand scheduler with an MCP server, publishing across eight platforms
- Wrote scheduled jobs on a Mac mini that draft content, check quality, schedule posts and draft replies
- Added deterministic quality gates and approval boundaries so nothing public ships without an OK
- Moved the stack to Codex first, with local models as a fallback, to control cost
- Built self-healing checks and daily receipts so failures are visible rather than silent
What it shows
- Practical AI engineering: orchestration, evaluation and guardrails
- Systems thinking across many small services
- Designing automation that people can trust and audit
Need this on your team?
I take contract roles as a senior design or product engineer, and I'm open to permanent.