Knowledge podcasts
You give it a topic, you get a researched podcast episode — and by now an explainer video to go with it. It started with episodes for kids; a second system for adults grew out of that: background episodes on heat pumps, legislation or inflation. Every statement is checked against sources before a voice speaks it.
- For
- Kids (6–12) & adults
- Per episode
- ~10 min podcast + ~4 min video
- Checked
- 56–98 statements per episode
- Status
- Prototype, in trial
Background episodes that explain instead of oversimplifying
A topic, optionally with my own links or PDFs. Out comes a research dossier, a script for one or two voices, a podcast episode and a 16:9 video with subtitles. The clip is a 38-second excerpt from the heat pump episode.
Three looks, same pipeline
The look is chosen per episode; the vertical version for phones is rendered alongside. Illustrations come from an image model; numbers, charts and lists are drawn by the pipeline so they stay exactly right — with the source in small print.
Two voices, one asks the questions
The kids' episodes are dialogues: one voice explains, the other asks what kids would ask. The video has clay-style pictures, a thermometer, maps and a quiz at the end. The clip shows the first 37 seconds of “Fire mountains” (volcanoes).
Six real runs
From the October 2026 test series. No statement was rated wrong; “minor” means a phrasing that should be more precise — those I resolve before approval.
| Episode | Sources | Statements checked | Minor | Podcast | Video |
|---|---|---|---|---|---|
| Heat pump (comic) | 45 | 63 | 1 | 9:30 | 4:00 |
| How a law is made | 44 | 83 | 1 | 9:40 | 3:50 |
| Inflation | 51 | 98 | 5 | 13:20 | 3:50 |
| Battery storage | 43 | 95 | 2 | 13:40 | 3:50 |
| EU deforestation regulation | 36 | 68 | 4 | 11:30 | 4:00 |
| Volcanoes (kids) | 47 | 56 | 0 | 11:40 | 11:40 |
Trustworthy first, pretty second
Sourced, not made up
Every statement in the script is checked against the collected sources. Anything that doesn't hold up gets rewritten before a voice speaks it.
Understandable
A second check looks for technical terms that aren't explained — especially important for the kids' episodes.
Approval before production
Before audio and video are produced, I read the script in the editorial studio and approve it — or send it back.
Costs in view
Every AI call is logged with its price. A complete episode — research, podcast, video — currently costs about $2.50–4.50.
Selectable looks
Comic, editorial, isometric or clay — chosen per episode. Diagrams and charts are drawn by the pipeline itself, not by the image model.
One script, two formats
Podcast and video come from the same script. Slides and subtitles follow the spoken word; the video also exists in 9:16 for phones.
Roughly how it works
01
Topic
A question — optionally with my own links or PDFs.
02
Research
Collect sources, extract and verify facts.
03
Script & check
Solo or duo script, fact check, comprehensibility.
04
Approval
I read it in the studio and approve it.
05
Podcast + video
Voices, images, slides — one episode, two formats.
Under the hood: TypeScript with Mastra workflows, language models from OpenAI, voices from ElevenLabs, image and video models via Replicate, PostgreSQL and a job queue. The editorial studio is a Next.js app.
Frequently asked questions
What comes out of a topic?
A research dossier with sources, a script for one or two voices, a podcast episode (about 10–14 minutes) and an explainer video with subtitles in 16:9 and 9:16.
How do you keep the AI from making up facts?
Every statement in the script is checked against the collected sources. In the test series that was 56 to 98 statements per episode; none was rated wrong. Before production I read the script and approve it.
Which AI models are used?
Language models from OpenAI for research, script and checks, ElevenLabs for the voices, and image and video models via Replicate. Charts and numbers are drawn by the pipeline itself.
What does an episode cost?
A complete episode with research, podcast and video currently costs about $2.50–4.50 in AI costs. Every call is logged with its price.