Skip to content
G
← All projects

Image Generation Platform

Async AI image generation on Kubernetes, with similarity search

A platform that generates images with a generative AI model at scale. Requests go to an async API that answers immediately with a job ID; Redis queues the work and tracks which generation servers are free, so each job is allocated to an available worker running as a pod on Kubernetes. Every finished image is embedded and pushed into a vector database, making the whole library searchable by visual similarity.

My role
Architecture, async API, job scheduling & Kubernetes deployment
Stack
KubernetesContainersRedisAsync APIGenerative AI modelEmbeddingsVector database

Architecture

submitenqueueassignembedsearchClient appsSubmit · poll · searchAsync APIJobs · status · searchRedisQueue · allocationGenerator pod AK8s · image modelGenerator pod BK8s · image modelGenerator pod CK8s · image modelVector databaseImage embeddings
overview

Highlights

  • Async by design: the API never blocks on a slow generation — clients get a job ID instantly and poll for the result.
  • Redis-based resource management allocates each job to a free generation server, so expensive model capacity stays busy without being overloaded.
  • Generation servers run as Kubernetes pods: self-healing, consistently deployed, and able to grow with demand.
  • Every image is embedded into a vector database, turning a pile of generated files into a library you can search by similarity.

Curious how this would fit your team?

Ask me anything about Image Generation Platform — I reply personally.