Deepfake detection as a service, on Kubernetes

The problem

Datambit needed to offer deepfake detection to enterprise customers who upload media in bursts: nothing for hours, then thousands of files at once. Detection models are heavy, so a single service would either sit idle or fall over.

What we built

A microservices platform where each media type has its own detection service, jobs flow through RabbitMQ, and Kubernetes scales pods horizontally and vertically with the queue. A Next.js web front end and a Kotlin mobile client submit media and track results. The architecture, deployment and both clients were delivered as one piece of work.

What it changed

Capacity follows demand instead of being provisioned for the peak. New detection models slot in as new services without touching the rest, and enterprise customers get a consistent interface across web and mobile.