Presentation: SafeChat: Building AI-Powered Safety Systems at Scale in a Real-Time Marketplace
Bruna Pereira explains how DoorDash built a content-agnostic AI moderation platform. She covers replacing costly LLM-only pipelines with a hybrid pattern: using fast internal models to filter obvious cases, LLM multi-axis scoring for nuanced decisions, and no-code workflows with backtesting. Discover how this architectural pattern cut safety incidents while scaling to millions of daily messages.…
InfoQ presents "SafeChat: Building AI-Powered Safety Systems at Scale in a Real-Time Marketplace" by Bruna Pereira, a software engineer at DoorDash with over a decade of experience in software engineering. Bruna discusses how DoorDash created a hybrid moderation platform to tackle safety issues in real-time conversations. The system uses fast internal models to filter obvious cases, LLM multi-axis scoring for nuanced decisions, and no-code workflows with backtesting.
Bruna explains how this architecture reduced safety incidents while scaling to millions of daily messages. DoorDash is a marketplace connecting consumers, Dashers, and merchants, with chat and voice being primary communication channels. Ensuring the safety of these interactions is crucial, with over 4 million chat messages exchanged daily.
Initially, DoorDash considered using an LLM-only pipeline, but this would have been too costly and slow. Instead, they adopted a hybrid approach, starting with a fast classifier to identify obviously safe messages. When the classifier couldn't determine if a message was safe, it was passed to an LLM for further analysis. This approach significantly reduced costs and latency while maintaining high safety standards.
Written by urgent.news from InfoQ's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
