The collaborative research project KISEE (AI-based Skill Assessment and Development) explores the design possibilities of an interactive AI avatar to enhance access to future skills, particularly for individuals from educationally disadvantaged backgrounds. The avatar aims not only to complement traditional counseling approaches, such as chatbots or employment agencies, but also to make them more accessible, scalable, and effective. By providing personalized guidance, the avatar assesses existing competencies for identifying new career opportunities and facilitates access to training programs and certifications.

As part of the KISEE project, a systematic literature review of 33 academic studies was conducted to identify key factors influencing avatar design. The findings highlight the following aspects:

Gender and Stereotypes:
Avatars are subject to gendered perceptions. While female avatars are often associated with assistant roles, masculine avatars can increase interaction intensity. A gender-neutral design might help avoid stereotypes, but users tend to unconsciously assign gender based on contextual cues such as conversation topics or linguistic features.

Avatar Design and the Uncanny Valley:
Striking a balance between realism and stylization is crucial to avoiding the uncanny valley effect. Human-like avatars are generally perceived as more trustworthy, whereas stylized avatars elicit more positive emotional responses and may reduce cognitive load.

Trustworthiness:
The acceptance of an AI avatar strongly depends on transparency and clearly defined roles. Users prefer avatars in supportive rather than social roles and generally trust real humans more. A carefully balanced design—realistic yet visibly artificial—can maximize both trust and acceptance.

These findings provide valuable insights into designing effective, trustworthy, and inclusive AI avatars that promote equitable access to future skills and sustainable professional development.

Keywords: AI Avatar, Skill Assessment, Career Guidance, Acceptance, Human-Computer Interaction.