Podcast Production — Proof of Concept
Technical prototype portfolio — demonstration environment
This public demo runs only the bundled synthetic sample episode below. Real multi-speaker text-to-speech (espeak-ng, a genuine system synthesizer — no AI call) is stitched into one real MP3 with real embedded chapter markers. The output is independently re-read with ffprobe afterward and checked against what was actually declared, not trusted because the encode step didn't error.
Episode title
Verifying AI Output, Not Trusting It
Synthetic sample script
Welcome back to the show. Today we're talking about how to verify AI output instead of just trusting it.
Right, and the failure mode we keep seeing is a pipeline reporting success when nobody actually reread the real output.
Thanks for having me. The short version is: never trust the pipeline's own report that it worked. Re-read the real output and check it against what you actually asked for.
That wraps up this episode. See you next time.