Customer Service
Lyft self-serve customer support agents
Lyft uses router-based LangGraph agents to route rider and driver support issues across configurable subagents.
Author: Akshay Sharma
What was done
Lyft's SCX Data Science and MLE team built a self-serve customer support platform where domain experts define configurable agents with prompts and JSON config. A LangGraph meta-agent routes rider and driver conversations to specialized subagents for account access, damage claims, charge reviews, earnings disputes, and other workflows. LangSmith traces, dashboards, and LLM-as-judge evaluations support production monitoring, while DynamoDB persists multi-turn state.
Stack
LangGraph
LangSmith
DynamoDB