Wave Forge
Audio conversion as four small services talking over a queue.
4
services in Go, Node.js and Python, coordinated over RabbitMQ
- Images on Docker Hub
- Written for Kubernetes
- Go, Node.js and Python
The problem
AI training data needs audio in consistent formats and sample rates, and converting it is slow work that shouldn't block an upload.
What I did
I built all four services and their Docker images, wrote the Kubernetes manifests, and ran the gateway on Koyeb with the other services on AWS EC2.
How it works
Gateway
Validates the request and the user's token, then hands the job off.
Auth
Handles sign-in with Clerk and tracks each user's files.
Converter
Picks jobs off RabbitMQ and converts format and sample rate.
Notification
Emails the user through Mailgun when the file is ready.
Next project