I made a simple workflow to transcribe podcasts. It lets me consume podcast content in my most enjoyable and speedy manner.
-
serverless GPU on Runpod.io - ~$0.15 and 1 minute to transcribe 60 minutes of audio. Open source repo, one-click deployment
-
a job runner on Cloudflare KV and S2 free plans. It took me ~6 hours to make with Cursor. I had templates, and most of time was spent on podcast data extraction quirks and plumbing. Inference works fine off the shelf
-
a button to "Copy Raw Transcript" which I paste in Z.ai web chat. A simple prompt[^1] (plus a lot of raw text!) immediately starts producing a pleasing and readable transcript in the browser
speaker identification, with chapter headings
audio file to raw text, cheap and fast
On Manifold Podcast, 9 Oct 2025, theoretical physicist and entrepreneur Steve Hsu spoke with Zixuan Li of Z.ai.
Li Zixuan graduated from Tsinghua, and got Masters from MIT and Carnegie Mellon. He currently works as Head of Product for Z dot AI, the...