The Brief
In 2016, running FFmpeg inside AWS Lambda was an odd idea. The runtime was constrained, binaries had to be packaged by hand, and most teams defaulted to always-on servers. A client with a media product wanted encoding that cost nothing while idle and scaled instantly under load.
What I Built
- FFmpeg-on-Lambda encoding: packaged binaries, tuned memory/duration trade-offs, S3 in, S3 out, API Gateway in front.
- An EC2 overflow tier for jobs beyond Lambda's limits, scaled up and torn down automatically.
- Successive revisions across six contracts as formats, volumes and AWS itself changed, December 2016 to February 2018, then a seventh contract for their Laravel application on the same AWS account.
- Adjacent serverless work the pattern unlocked: ImageMagick GIF pipelines, Python ETL, DynamoDB backends, Zoho integrations.
7contracts with the same client on this pipeline
5 yrsrelationship span, 2016-2021
20+AWS contracts delivered since 2016
$0idle cost. The point of going serverless
Why It Holds Up
Media infrastructure earns trust by being boring: predictable costs, observable failures, retries that work. The measure of this system is not a benchmark. It is a client who came back six times in fourteen months to grow it.