AI Customer-Service Agents · E-commerce (Shopify)
Customer-Service Agents for E-commerce
Voice and chat customer-service agents for a Shopify e-commerce store, built on LiveKit and LangGraph, traced with LangSmith and running on AWS. The client is confidential and is not named here.
The challenge
A Shopify e-commerce merchant — whose name we do not publish, by agreement — wanted customer service answered in conversation rather than queued as tickets: a shopper asking about an order, a return or a product should get an answer by voice or by chat. A scripted bot was not enough. The agents had to follow multi-step flows, hold state across a conversation, hand off when a human is needed, and be observable enough that the team can see exactly what an agent said and why.
What we built
We built the conversation layer on LiveKit, so the voice channel is real-time instead of a recording sent back and forth, and wrote the agent logic in LangGraph: each flow is an explicit graph of steps with state, not a prompt chain nobody can audit. LangSmith traces every run — each step, each tool call, each answer — so a wrong answer is inspected rather than guessed at. The whole system runs on AWS.
What changed
Customer-service agents that talk with shoppers by voice and by chat over the store's own Shopify catalogue and orders: LiveKit for the live conversation, LangGraph for the agent flows, LangSmith for tracing and evaluation, AWS for the infrastructure. The engagement is covered by a confidentiality agreement, so the client is not named and no operating figures are published here.