Backend Engineering for a Podcast Transcription PlatformBackend Engineering for a Podcast Transcription Platform
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
The product
Paste a podcast link from Apple Podcasts, Spotify or RSS and get a transcript, summary, chapters and FAQs. Transcripts go into a public, searchable library, and developers can use the same engine through a paid API.
My role
I took over in September 2025 and have written about 576 of the 593 backend commits since, including the move to self-hosted transcription.
What I built
Replaced a paid speech-to-text API with whisper.cpp on our own GPU server
A job queue across two servers with paid, free and crawler tiers and crash recovery
Long-audio processing in chunks, with correct timestamps and real progress
AI summaries using a local and a hosted model, under a daily spending cap
A crawler that transcribes back catalogues while the GPU is idle
Speaker labels, search, chat with an episode, PDF/SRT/VTT and Notion/Obsidian export
A public API with credit plans, and a second brand running on the same backend
Hard problems
A clean-up step that froze the server: removing Whisper's repeated lines was quadratic and blocked the server for minutes on long episodes. I rewrote it to run in bounded time.
Billing that's right every time: API refunds are claimed exactly once, and an append-only ledger records every credit change.
Results
52,700+ public transcripts
41,000+ hours of audio transcribed
73 languages
3,200+ backend tests
Post image
Back to feed
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started