Why the AI character would not calm down, and how I fixed it

Description and use

The AI character in Say It Ahead remained upset despite user attempts, making practice arbitrary. Implementing a four‑job model and a live progress tool enabled the character to gradually calm down and provide clear visual feedback.

Problem

The AI character in Say It Ahead remained upset despite user attempts to calm it, making practice feel arbitrary and ineffective.

Outcome

The character now adapts its tone based on user actions, providing a clearer sense of progress and reducing arbitrary outcomes.

Workflow

• Define scenario with character mood, beliefs, facts, and desired resolution.
• Structure the conversation around four jobs: acknowledge, clarify, reflect, move forward.
• During the call, the AI calls update_practice_progress to report the highest job completed and evidence from user input.
• The UI displays progress silently, giving users visual feedback without breaking role‑play.
• After the call, ElevenLabs runs a separate review to score each job and maps results to the review screen.
• Compare live progress with post‑call review to identify discrepancies and refine criteria.
• Iterate tests with varied wording to reduce variation and improve consistency.
• Deploy updated model in production.

Evidence

The live character can use a browser tool called update_practice_progress. It sends the highest job completed and a short piece of evidence from what the user actually said.

Source

The current guided scenarios use four broad jobs: acknowledge the concern, ask a useful question, reflect what you heard, and agree on a specific next step.

Source