Before
Video production is not one service. Autovid had APIs accepting work, workers processing media with FFmpeg, open-source video-production systems, AI jobs, and Temporal coordinating long-running workflows. Kubernetes scheduled those containers, but operating Kubernetes became another system the team had to build around. Every new component brought more platform work: manifests, service configuration, secrets, deployment automation, logs, metrics, and the glue required to make one workload discover another. The team could run the software, but each release carried infrastructure work that was separate from improving the video product.Kubernetes plumbing
The team maintained the deployment and operational layer around every workload.
Video processing
FFmpeg and open-source video systems handled media work across many workers.
AI workflows
AI processing added more long-running jobs and services to operate.
Temporal
Temporal coordinated workflows, while the team maintained the services around it.
What moved
Autovid did not need Springwinter to translate a Kubernetes cluster. The team deployed the systems it already used as containers: web services for APIs, workers for FFmpeg and AI processing, open-source video-production components, and the services and workers used by Temporal. Web servers and workers build from GitHub and run as ECS services in Autovid’s AWS account. Supporting databases, caches, and storage stay in the same account. The workload remains made of the open-source tools Autovid chose; Springwinter handles the repeated path from repository to running AWS resource.Less platform work
A deployment now starts from the repository instead of a set of cluster-specific release steps. Springwinter runs the build, updates the service, and keeps the deployment record together with its output. A branch can run as a preview before replacing a live service. Logs and metrics come from the AWS resources where the workloads run. Cost is visible beside those resources. The team no longer has to assemble a separate operational path for every FFmpeg worker, AI process, or open-source service it adds.1
Keep the workload
Continue using FFmpeg, open-source video-production software, AI workers, and Temporal as containerized systems.
2
Leave the cluster plumbing
Replace Kubernetes-specific deployment and operational work with Springwinter resources in the existing AWS account.
3
Deploy from GitHub
Build and update web services and workers from their repositories, with previews available for branch changes.
4
Operate from one project
Read deployments, logs, metrics, and cost beside the resources that produce them.
The result
Autovid reduced infrastructure cost by 10% without materially affecting how its production workloads behaved or their reliability. Removing deployment plumbing also changed how quickly the team could ship: it now makes 3x as many deployments as it did on Kubernetes. The gain did not come from replacing FFmpeg, Temporal, or the open-source video stack. It came from keeping those systems while removing the platform work that had accumulated around them.10% lower cost
Infrastructure cost fell without a material service impact.
3x more deployments
The team deploys three times as often as it did before.
Workers
Run FFmpeg, AI processing, and workflow workers in the AWS account.
Logs and metrics
Read workload output and operating signals without building another system.