03 / CLIENT PROJECT
AI voice agent evaluation
A client project via Upwork, evaluating a Slovak-speaking AI customer-service agent through realistic phone conversations.
The assignment
The client’s brief called for two realistic customer-service phone conversations with an AI voice agent: one airline scenario and one healthcare scenario, followed by an evaluation of each in an annotation platform.
The review criteria covered communication quality, response accuracy, conversation flow and the overall experience. As a native Slovak speaker, I could also assess how understandable and natural the spoken interaction felt.
What I observed
Clear communication
The conversation was smooth, clear and easy to understand. The agent responded well to my words and requests.
A robotic voice
Although the speech was understandable, the voice sounded too robotic. Clarity and naturalness were distinct aspects of the experience.
Pauses that felt too long
The pauses felt unnecessarily long and affected the pace of the interaction. This case study reports a qualitative observation rather than measured response times.
The key finding
The agent handled the interaction well, while the voice and pacing left room for improvement. A useful evaluation needs to capture both what works and what makes the experience feel less natural.
What this work demonstrates
Evaluating a conversational agent means paying attention to more than whether it produces a response. I distinguished how well it responded to requests from how its voice and timing felt during a real conversation.
Scope of this case study
This is a qualitative account of my experience on a small client assignment. It does not establish the agent’s overall accuracy or performance, and it makes no claim about clinical accuracy or changes made after the evaluation.
The client is unnamed. Call recordings, phone numbers, conversation transcripts and proprietary annotation materials are not included.