Shane Harrell: Demo: cat support_engineering | uniq

8/12/2026

10:45 am

-

11:05 am

Speakers

Shane Harrell

Add to Calendar

About the Session

Support at SignalWire has developed a platform integrated with AI that allows for unparalleled data retrieval and analysis which greatly reducing response times for our customers.

‍

Add to Calendar

Related Sessions

View All

Demo

Brian West: Every voice AI demo ends at deployment. This one starts after the call.

In this live demo, Brian West shows how SignalWire’s post-conversation payload turns a completed Voice AI session into a full operational timeline. Using real open-source demos like GoAir and Holy Guacamole, he walks through the JSON payload generated at the end of a call and shows how it can be used to inspect transcripts, response timing, ASR, LLM, and TTS latency, barge-in behavior, SWAIG function calls, global_data snapshots, and state changes across the entire conversation. The demo also introduces the open-source PostPrompt Viewer, a visual tool for analyzing completed Voice AI calls. Attendees will see how teams can turn raw post-call payloads into practical observability metrics like P95 latency, barge-in rate, tool execution overhead, and response verbosity. A transcript tells you what happened on a call. SignalWire’s post-conversation payload helps show why it happened, where time was spent, and how the AI moved through the conversation. This is what carrier-grade Voice AI observability looks like after deployment.

10:45 am

-

11:05 am

Demo

Justin Massey: Demo - Building an AI voice agent that answers real customer calls while staying provably on-script.

LLMs are great conversationalists, but can be terrible rule-followers. Point one at a customer call and it'll happily chat, but skip the questions you actually need answered, ask them out of order, or invent a path through your call flow you never intended. Old-school IVR solved determinism by removing the conversation entirely. That's not the trade we want to make. In this live demo, I'll build an AI voice agent that answers real customer calls while staying provably on-script. You'll see how to constrain an LLM so it collects every required piece of information, asks the right questions at the right moments, and honors your call tree, without collapsing back into a rigid "press 1 for support" menu. The agent still improvises, handles interruptions, and sounds human;it just does so inside guardrails you define.

3:30 pm

-

3:50 pm