Senior Software Engineer - Roblox
Scale AI
Audio
ClassificationLed end-to-end development of real-time AI-powered content moderation pipelines, deploying transformer-based NLP models for detecting policy violations in chat and voice at multi-billion event scale. Architected and launched a fault-tolerant microservices infrastructure on Kubernetes with GPU-accelerated inference using Triton and TorchServe, ensuring sub-100ms latency and 94% uptime during global traffic spikes. Improved moderation accuracy by 15%, reducing dependency on human review through ensemble modeling of multilingual filters and rule-based validators. Spearheaded integration of generative AI tooling into Roblox Studio, launching the Roblox Assistant - an LLM-powered natural language coding assistant to help creators generate Lua scripts and 3D assets via prompts. Fine-tuned and deployed LLaMA and CodeGen models for in-editor code generation, reducing prototyping time by 40% and increasing adoption among novice creators. Showcased Roblox Assistant at RDC 2025, demonstrating full-stack integration of LLM workflows with user telemetry and in-editor feedback loops. Built low-latency audio ingestion and classification services using gRPC and containerized streaming APIs, cutting moderation response time by 20% and improving user safety across voice-enabled experiences. Conducted rigorous A/B testing on moderation recall vs latency trade-offs, and integrated a feedback-driven retraining loop for model updates based on flagged events. Collaborated with product, safety, and trust & security teams to align ML infrastructure with Roblox’s Responsible AI principles, driving strategic investments in content safety and developer enablement.