The Roblox PII Classifier goes beyond detecting and obfuscating explicit PII text shared on Roblox, it’s designed to understand the context of a chat conversation to stop bad actors from engaging in PII-related conversations in the first place.
One of Sentinel’s key advantages is that it does not require a large number of exemplars to function. Our current production system operates successfully with just 13,000 exemplars in the negative index.
Roblox’s LLM is currently outperforming popular LLM guardrail models on standard benchmarks. Roblox open-sourced both the LLM weights and the RoGuard-Eval benchmarking dataset.
At the heart of our engineering productivity strategy are three tools: our microservice lifecycle platform, our code center—an inner loop development tool—and our advanced observability platform. Together, these tools enable more than a thousand Roblox engineers to tackle challenging problems
