Project · Live

Peace AI Briefing

New AI papers in, fact-checked explainer videos out every day, presented by a custom-built avatar and improved by its own analytics.

Watch on YouTube

Architecture

Architecture diagram: sources and ranking feed a LangGraph script agent with write, policy gate, fact-check and human approval; then render with a custom avatar engine, publish to YouTube, and an analytics agent feeds learnings back to the script agent.

Click the diagram to open it full size.

The six stages

1Sources

Trending AI papers (arXiv, Hugging Face) and big-tech AI announcements, collected daily.

2Rank & plan

Local embeddings rank and de-duplicate papers; paper figures are used only when their licence allows it.

3LangGraph script agent

Write → policy gate → fact-check against text and figures, with a rewrite loop on failure, checkpoints, and human approval for new formats.

4Render

Local text-to-speech, Whisper alignment, figure slides and subtitles in five languages, with the in-house avatar layered on top.

5Publish

YouTube Data API: video, thumbnail, playlist, captions and chapters, with sources and an AI-assistance note in every description.

6Data analysis

Retention per sentence, 3-second hook and format comparisons; every number is re-checked against the raw data, and findings become writing rules.