Supplier conversations that can be inspected and improved
An agent that talks to suppliers is only useful if you can see why a conversation went the way it did, and fix it.
We rebuilt the supplier-conversation agent around explicit goals, conversation history and human escalation, with end-to-end tracing and regression tests behind it. Early conversations showed that better replies drew more complex supplier questions. That changed which failure cases to handle before a wider rollout: for example, the agent has to stop pushing once a supplier declines.
R&D Lead · agent implementation and evaluation. Teammates handled messaging transport and infrastructure, integration and product input.
Tested against real supplier conversations, with failure cases documented.