# AI Event Co-host System

Client: National Instruments Leadership Forum
Canonical path: /case-studies/ai-event-cohost

## Summary

A live AI voice assistant built in four days to co-host an enterprise leadership forum with scripted segments, listening mode, waveform monitoring, and guarded responses.

## Outcome

84% line accuracy during the live flow and 94% attendee usefulness rating.

## Challenge

The event needed an AI co-host that could stay on script, respond to cues, and feel polished in front of an enterprise audience without creating live hallucination risk.

## Evidence

- Build window: 4 days
- Live line accuracy: 84%
- Attendee usefulness: 94%
- Primary risk controlled: Off-script responses

## Approach

1. Built a controlled script-and-cue interface instead of a free-form chatbot.
2. Added listening mode, waveform monitoring, and guarded response states so the operator could see when the system was active.
3. Prepared scripted fallback responses for predictable event transitions and sponsor moments.
4. Kept the live system narrow: co-hosting and Q&A support, not open-ended enterprise advice.

## Timeline

1. Scope and script
2. Voice/response prototype
3. Operator interface
4. Live rehearsal
5. Event run

## Risks

- Audio routing failure
- Hallucinated sponsor detail
- Latency during handoff
- Operator overload

## Stack

- Next.js serverless app
- OpenAI response pipeline
- Audio waveform visualization
- Custom listening mode

## Next Evidence Step

Replace provisional metrics with client-approved screenshots, raw artifacts, and final numbers when available.