2025 · Platform
A presigned upload pipeline for a custom GPU model
a media processing startup · Cloud & ML infrastructure engineer
The problem
A client had a ready model in its own container but no way to receive large uploads and pass them to it. Sending files through a web server does not suit video and large images.
What I built
I built the cloud setup: a small function that hands out short-lived upload links, storage for the files, sign-in for callers, and a GPU container service that runs the client's model.
The result
The client could take real uploads and run its own model on them, with no upload server in the middle.
Main tools
- AWS Lambda
- Amazon S3
- Amazon Cognito
- Amazon ECS
- Terraform
This is the short version. The full case study is available on request.