docs/demo_script.md

# Demo Script

## 2-Minute Version

I built a forward-deployed AI simulation to mirror how Distyl operates. Instead of starting with a model, I started with an ambiguous enterprise problem: massive noisy support tickets and fragmented workflows make it hard to explain VIP churn drivers. I defined success metrics first, mapped the current workflow, and explicitly decided where AI should vs should not be used. Then I rapidly prototyped a system that turns unstructured tickets, emails, and chat conversations into auditable structured outputs with evidence citations, risk gating, and human review routing. Finally, I built an evaluation harness, documented failure modes, and abstracted the solution into reusable modules so the pattern generalizes to other enterprise workflows.

## 5-Minute Version

*To be finalized after system is built and measured.*

## Key Demo Flow

1. **Problem Scoping** — Show the AI suitability matrix and explicit non-goals
2. **Prototype Lab** — Process a case bundle end-to-end, inspect structured output
3. **Reliability & Review** — Show risk gate routing a high-risk case to review queue
4. **Abstraction Layer** — Walk through reusable modules and adjacent use cases