Backend & AI Engineer
Whisper-Based Audio Transcription Service
Context
Client engagement.
A company relied on manual transcription of recorded audio. A single file could consume hours of a user's day, which limited throughput to roughly one completed transcription per person per day.
Engineering approach
I built and deployed a focused transcription service instead of presenting AI as a general-purpose feature.
What I owned:
- Implemented the speech-to-text workflow in Python using OpenAI Whisper.
- Containerized the service with Docker so the same runtime could be tested and deployed consistently.
- Integrated the service into the existing product so users could submit recorded files and retrieve the generated transcript through a repeatable workflow.
- Kept model execution behind the application boundary rather than exposing provider or infrastructure details to the client.
Outcome
Reduced a workflow that previously consumed hours to minutes and increased user throughput from approximately one transcription per day to more than ten.