Blogs›Lyft›Analytics & Rides Intelligence Assistant (ARIA)

Analytics & Rides Intelligence Assistant (ARIA)

posts

2026

Lyft built and launched the production frontend for its AI-powered Analytics & Rides Intelligence Assistant (ARIA), enabling employees to explore ride analytics data through natural language. This involved migrating from a prototype Streamlit interface to a production-grade web client using Node.js and Next.js, integrating with Lyft's authentication, deployment infrastructure, and observability tools. The project also involved establishing Server-Sent Events (SSE) for real-time data streaming and developing a framework for estimating long-term effects of marketplace decisions using surrogacy and region-splits, incorporating observational causal inference, residualized regressions, and switch-back/user-split experiments for validation.

2026

Beyond A/B Testing: Using Surrogacy and Region-Splits to Measure Long-Term Effects in Marketplaces

7/21/2026

This post details a framework for estimating long-term effects of marketplace decisions, specifically focusing on 'market-mediated long-term effects'. It introduces a two-step approach: 1) estimating how policy changes affect negative user experiences (e.g., long waits, high surge) using residualized regressions and validating with switch-back experiments, and 2) estimating how these negative experiences affect future user behavior using double-robust causal inference (AIPW) and validating with user-split experiments. The overall long-term effects are then verified using region-split experiments, with a forward selection algorithm for optimizing region selection.

From Day 1 to Production: Building Lyft’s Analytics & Rides Intelligence Assistant as Onboarding…

7/10/2026

This post details the process of building the production frontend for Lyft's ARIA system as an onboarding project. It covers the migration from a Streamlit prototype to a Node.js/Next.js application, including setting up the repository, configuring the Node.js service scaffold with authentication, logging, and metrics plugins. It describes the infrastructure work involving Envoy static configuration, DNS setup, and staging deployments, highlighting the resolution of an authentication plugin incompatibility. The post also details the connectivity layer to the backend, the use of Grafana for debugging Envoy configuration issues, and the implementation of an XState state machine for chat state management. Finally, it covers the integration of Server-Sent Events (SSE) for streaming responses and production hardening steps.