How it is made
Each episode is produced by a scheduled, automated pipeline. The daily run is scheduled for 06:00 East Africa Time (03:00 UTC). Scheduled runs can start later than their scheduled time, so an episode may appear after 06:00. The scheduled workflow does not include a manual approval step between collection and publication.
The daily run
-
Collect
Headlines and item details are read from public Google News search feeds for each section. The day’s weather for Nairobi is requested from the wttr.in weather service. The pipeline does not open the articles that the feed items link to.
-
Select and write
An AI language model (OpenAI, model gpt-5) applies a written editorial prompt to the collected material. It decides which stories to cover and how to present them, and writes the spoken briefing. Each day’s headlines are saved so the next run can avoid repeating them.
-
Narrate
The written briefing is read aloud with OpenAI’s text-to-speech model (gpt-4o-mini-tts, voice “alloy”).
-
Build the episode
The programme’s own renderer draws the video from the briefing’s segments, with captions, lower-third titles, and map and text graphics. Map geometry comes from Natural Earth data, which is in the public domain. The renderer does not use photographs or AI-generated images. An episode thumbnail is also produced.
-
Deliver
The briefing text and audio are sent to a Telegram chat. The video is uploaded to YouTube and published publicly.
What is AI here
- The written briefing is produced by an AI model, working under written editorial instructions.
- The narration is an AI-generated voice.
- The graphics are drawn by the programme. They are not generated by AI, and the current version uses no photographs.
Limits
- Briefings can be wrong or incomplete, because they are written from feed items rather than from the full articles.
- The scheduled run publishes without a human approval step.
- This website does not host audio or video files. Episodes are delivered through Telegram and YouTube.