The stack
Tool by tool, no omissions.
The "AI-Dima" avatar is generated using HeyGen. I record a script read-through, HeyGen produces the synced video at scale.
Why:One human, many languages, consistent delivery. A Ukrainian-born educator can talk to Indonesia, Vietnam, and Thailand simultaneously.
Honest: The face is generated. The voice is AI-synced. The script is written by me.
Every script goes through Claude. Research, outline, draft, refine. I prompt with specific constraints — audience, tone, what not to say.
Why:One operator brain, AI production speed. Claude produces the first draft; I cut, rewrite, and make the editorial calls.
Honest: First draft is AI. Final voice is mine. I review everything before it goes out.
Content scheduling, cross-platform posting, and data pipelines run through n8n. No manual uploads to Instagram or Telegram.
Why:Consistency without manual overhead. n8n lets a solo operator run a media cadence that would require a team.
Honest: No manual scheduling. But I decide what goes into the pipeline — n8n doesn't pick topics.
The zapleo.com site — including this page — is written with Claude Code (Anthropic's coding AI). I describe what I need; it generates the code.
Why:I am not a full-stack developer. Claude Code lets me ship a production-quality site without a dev team.
Honest: Code is AI-generated. Architecture decisions, content strategy, and final review are mine.
The principles
Why I do it this way.
Disclose everything
If AI touched it, say so. An AI educator who hides the AI is teaching the wrong lesson. The avatar is generated. Say it in the video. The script is AI-drafted. Put it in the description.
The human decides what to teach
AI doesn't pick topics. It doesn't choose what to recommend or what to omit. The editorial judgment — what's worth teaching, who to trust, what advice is dangerous — that's the job that doesn't get automated.
Production is the cheap part
The hard work is staying current, interviewing employers, reading actual job postings, and filtering bad advice from good. AI handles the production; the human handles the knowledge work.
This is the model I teach
Every student of AI-Dima learns this split: use AI for production, keep humans in the decision seat. I can't teach that while hiding how I work. So I show it.
Questions about the stack?
If you're building a similar AI-augmented project and want to compare notes, DM me. I'm genuinely interested in how other operators are running this.