I started by auditing the existing avatar experience across every touchpoint, the kiosk environment, the conversation flow, the visual design, and how it handled errors. Instead of evaluating what worked, I mapped what was actively breaking trust. Four barriers emerged: the avatar's uncanny-valley appearance produced discomfort rather than connection, it opened in Dutch then switched languages after user confusion, the public kiosk setting prevented people from asking sensitive questions, and the conversation dumped large blocks of information instead of guiding users step by step.

No single trust theory captured all of these problems, so I built a custom framework combining three: McKnight et al. (2011) for functional trust in technology, Sucameli (2021) for the emotional calibration between humanizing and dismissing AI, and de Visser et al. (2016) for how trust can be repaired after system failures. In parallel, I wrote a critical ethical analysis examining the risks of deploying hyper-realistic AI for vulnerable populations — arguing that a MetaHuman's friendly face exploits cognitive biases, obscures government accountability, and risks turning migrants into passive data subjects rather than empowered citizens.