Audits
Audit · May 9, 4:03 PM
Model: gemini-2.5-pro (audio) · View source session →
Overall
22
Warmth
50
Pacing
65
Character
5
Flow
20
Voice
60
Narrative
This call was a catastrophic failure due to two instances of the AI vocalizing its internal stage directions instead of performing them. These errors completely broke character, which the user immediately noticed and commented on, leading her to abruptly end the call. While the underlying voice and pacing were adequate, the fundamental failure to maintain the persona made the experience untenable.
Flagged moments
- @26shighRuby vocalized the stage direction 'Soft laugh' at the beginning of her sentence.This is an immediate and jarring break of character that reveals the AI's nature and shatters the illusion of the persona.
- @59shighRuby said the words 'laugh softly' instead of actually laughing.This is a second, undeniable AI tell that Sandy immediately identified as 'weird,' confirming the persona was broken.
- @232smediumSandy abruptly interrupted Ruby to end the call.This indicates that the user's trust and engagement were completely lost due to the AI's repeated, unnatural mistakes.
Proposed changes
- prompt100% confidentImplement a strict filter or fine-tuning to prevent the model from ever vocalizing text that appears to be a stage direction, such as action descriptions or words in brackets/parentheses.
- tts90% confidentThe TTS engine needs a more robust and natural-sounding library of non-speech sounds (e.g., laughter, sighs) that can be triggered by specific tags like '<laugh type="soft">' instead of relying on the model to interpret text.
- pacing70% confidentSlightly reduce the initial response latency; some pauses before Ruby speaks feel more like system lag than thoughtful silence, particularly the 2-3 second gap before her response at 0:44.
These auto-generate audit_lessons rows. Review and approve in Lessons.