Client: San Francisco, USA
VoicePatrol
Real-time audio moderation platform processing 10,000+ concurrent streams for VR gaming. Cloud-native architecture on Kubernetes and AWS.
Timeline: Multi-year partnership
Role: Platform + Core Engineering
Domain: Real-time voice safety
Challenge
Multiplayer VR sessions had escalating toxicity while moderation tooling lagged behind real-time voice interaction. The product needed deep audio analysis with strict sub-50ms reaction targets and global scale.
Solution
We shipped a hybrid pipeline with Rust services for critical stream processing and Python inference services for model execution, orchestrated on Kubernetes with autoscaling tuned for burst traffic.
Key decisions: WebRTC-native ingestion, deterministic buffering strategy, and multi-region failover.
Tech Stack
Results
1B+
Minutes Processed
99.99%
Uptime
<50ms
Detection Latency
“Jay Software owned the hardest part of our stack. They consistently shipped decisions that held up under real production load.”
— Product Leadership, VoicePatrol