← ulyssies.com
001
AI News Podcast Generator
A short audio news briefing on any topic, generated in seconds for the morning commute.
Live · In development
01 — Why I built this
I wanted an unbiased news briefing I could actually listen to on my commute, not another feed to scroll. Everything out there was either too long, too generic, paywalled, or shaped by a source's editorial slant. So I built one that generates a short, neutral audio podcast on any topic in seconds.
02 — How it works
01
Gather
Pulls 24-40 full-briefing articles from NewsAPI and Google News RSS.
02
Write
Claude turns raw articles into a tight, neutral briefing script.
03
Speak
OpenAI TTS converts the script into sentence-safe audio chunks.
04
Stream
Playback starts from the first chunk while the rest keeps loading.
ARCHITECTURE
Static web client, FastAPI pipeline, streaming audio path
Select a briefing to begin
03 — What was hard
- Streaming audio chunks instead of waiting for one finished fileThe first version generated a single audio file and waited for the full render before playback. Switching to sentence-safe chunked streaming lets audio synthesis overlap with script generation and starts playback before the episode is finished.
- Caching by topic and lengthWithout caching, every click to the same topic fired a fresh API call. In-memory caching makes repeat requests return instantly and avoids burning credits on duplicate content.
- Article volume by briefing lengthFull daily briefings pull from eight sections and target 24 articles for short, 32 for medium, and 40 for long before dedupe. That means the configured average full-briefing run uses 32 articles, while the 30-minute debrief targets up to 40.
04 — By the numbers
~5s
to first audio chunk
32
avg full-briefing articles
40
30m article target
05 — Links