Multi-agent AI · Content automation
Autonomous AI marketing crew with its own video studio
A team of AI agents led by an AI director that plans, writes, produces long-form video and publishes across social media — with compliance checks and a human in the loop only where it matters.
- 10
- specialised agent roles, coordinated by an AI director
- 5
- social platforms published to through one integration
- 400+
- test files across the agent backend and control dashboard
- 100%
- of outgoing content passes a compliance gate before publishing
01
The problem
A small team cannot produce daily, on-brand content in several formats and languages for five platforms — and in a regulated field every post and reply carries legal risk.
02
What I built
- An AI director that owns the content calendar, routes work to specialist agents and adjusts the plan from performance data
- Specialist agents: strategist, content drafter, curator, trend radar, creative producer, reply drafters, brand-voice linting and image compliance review
- Automated long-form video: script, AI-generated scenes (Veo 3), synthetic voices, subtitles (Whisper) and final assembly (ffmpeg), in several languages
- Publishing, comments and DMs across five platforms through the Ayrshare API; inbox agent that replies in-thread from a shared Gmail mailbox
- Discord as the control pane: approvals, asset requests, escalations and a full activity log
03
Engineering
- Typed, functional TypeScript (Effect) with a swappable model layer — Claude and Gemini tiered by cost and difficulty
- Risk-tiered autonomy: safe categories auto-publish, anything sensitive waits for human approval; every action is audited
- Idempotent inbound processing, rate-limited publishing and a separate web dashboard for operators