naman-kumar / journey / projects / netflix-game-launchPublic

Back to all projects

ShippedAKQA · for Netflix

It's What's Inside: The Game

50k+ concurrent users in the first 10 minutes

A Netflix game where up to six players impersonate each other using AI face and voice swapping. I led SRE for its launch, including the Kubernetes platform it ran on.

See the project on AKQA.com

My role
SRE lead
Game
Up to six players impersonate each other with AI face and voice swapping
Launch
50k+ concurrent users in the first 10 minutes
Platform
Kubernetes with GPU workloads, monitored in Grafana (NVIDIA DCGM)
A node-based AI workflow that animates a source portrait to follow the expressions of a driving video, with the movement input and the generated output side by side.
From AKQA's write-up: a face (right) animated to follow a movement input (left). Image: AKQA

Architecture

  1. Traffic

    • PlayersLaunch-day demand

  2. Platform

    • Kubernetes clustersDesigned for launch scale
    • GPU workloadsAI face transformation

  3. Observe

    • NVIDIA DCGMGPU metrics
    • GrafanaDashboards
Simplified platform view.

Problem

A public launch concentrates traffic into the first minutes, and the AI workload depends on GPUs, so the platform had to scale fast and GPU health had to be visible.

My contribution

  • Led SRE for the launch.
  • Designed the Kubernetes clusters that scaled to 50k+ concurrent users in the first 10 minutes.
  • Included Grafana monitoring of NVIDIA DCGM GPU metrics.

Engineering decisions

  • Kubernetes as the platform for the GPU-backed game workloads.
  • GPU health tracked through NVIDIA DCGM metrics in Grafana.

Reliability and security

  • GPU metrics (NVIDIA DCGM) monitored in Grafana.

Verified outcomes

  • Clusters scaled to 50k+ concurrent users in the first 10 minutes of launch.

Stack

Infrastructure is described at a high level; capacity figures beyond the public launch number are omitted.

git log project/netflix-game-launch: see this in the career journey