Sep 2025 – present
Podcast Transcript AI — transcription at scale
Took over and rebuilt Podcast Transcript AI: self-hosted Whisper on a GPU server and a public library of 52,700+ transcripts in 73 languages.
- Next.js
- Node.js
- Express
- MongoDB
Services · AI transcription app development
I build transcription and audio-AI products: self-hosted Whisper, job queues, long-audio processing, AI summaries with cost caps, APIs and payments.
Start a projectFounders and teams launching transcription, podcast, meeting-notes or media-search products.
Transcription looks simple until the bills and the edge cases arrive: two-hour files, queues that jam, AI summaries that quietly fail, and per-minute API prices that eat your margin. I build transcription products that stay fast and affordable as usage grows.
Sep 2025 – present
Took over and rebuilt Podcast Transcript AI: self-hosted Whisper on a GPU server and a public library of 52,700+ transcripts in 73 languages.
A call and a short written scope: what we're building, what success looks like, and what's out.
You see working software every week on a preview link, not a big reveal at the end.
I deploy to your servers or a VPS with monitoring, backups and HTTPS.
I stay on to fix, measure and improve — the same developer who built it.
It depends on volume. At low volume an API is simpler; once you transcribe thousands of hours, self-hosted Whisper on a GPU usually costs far less. I've done that migration and can model it for you.
Yes. Audio is split into chunks with their own timeouts and stitched back with correct timestamps, and uploads are processed from disk so large files don't exhaust memory.
Each feature routes to the cheapest model that does the job, with a daily spending cap and alerts, so a bug or a traffic spike can't run up a surprise bill.
Yes. I run two transcription brands on one backend with isolated accounts, sessions and payment keys.
Tell me what you're working on. I'll reply with questions or a first plan.
Start a project