← All work
RAG Product

Grounding a sales assistant in renewal knowledge

RAGWebexData governance
Delivered a GPT assistant for sales-renewal workflows.

The problem

Renewal work requires finding the right account and product context quickly. A generic chatbot can produce a plausible response, but plausibility is not enough when a sales team is acting on enterprise data and customer commitments.

What I shipped

A GPT chatbot with retrieval-augmented generation inside Webex for sales-renewal workflows. I led the cross-functional product effort, designed the retrieval architecture, and defined the data-governance guardrails end to end.

What was hard

Retrieval quality, access control, and user trust are coupled. Better recall is not a win if it crosses a data boundary, and a safe system is not useful if it cannot surface the context a seller needs. The product decision was how to make those constraints visible instead of hiding them behind a chat box.

What this means for AI products

RAG is not merely a technique for reducing hallucinations. In enterprise products it is the control plane for provenance, permissions, and the boundary between what a model knows generally and what a company authorizes it to use.

Running untrusted agents in production →Platform operations as agent-executable skills →Rules for the 80%, models for the tail →Right-sizing a fleet with a seasonal load curve →Shipping generative analytics before the playbook existed →Building the operational view from zero →Scaling content governance with policy and data →