Career Interview Readiness in Virtual Reality (CIRVR): A Platform for Simulated Interview Training for Autistic Individuals and Their Employers
Deeksha Adiani, Aaron Itzkovitz, Dayi Bian, Harrison Katz, Michael Breen, Spencer Hunt, Amy Swanson, Timothy J. Vogus, Joshua Wade, Nilanjan Sarkar · 2022 · ACM Transactions on Accessible Computing · doi:10.1145/3505560
Summary
This paper presents CIRVR (Career Interview Readiness in Virtual Reality), a closed-loop adaptive VR platform designed to help autistic individuals practice job interview skills. The system addresses a critical employment barrier: only 10-50% of autistic adults are employed, with job interviews being a major obstacle due to anxiety around open-ended questions, unexpected interruptions, and atypical eye contact patterns that may be misinterpreted by employers. CIRVR creates a realistic virtual office environment where users interact with an AI-powered interviewer avatar. The 12-15 minute interview simulation progresses through greetings, technical questions, education background, personal/behavioral questions, and a bidirectional segment where candidates can ask about the job. The system uses Microsoft Azure services for speech recognition, natural language understanding, sentiment analysis, and text-to-speech, allowing flexible conversation flow based on user responses. What distinguishes CIRVR is its multimodal sensing: an Empatica E4 wristband captures physiological signals (galvanic skin response, blood volume pulse) for real-time stress detection via a Random Forest classifier; Tobii eye trackers measure gaze patterns; and webcam-based facial analysis detects emotional states. This data feeds into a Dashboard designed for employers and job coaches to visualize how candidates experience interviews—not to penalize atypical behaviors, but to help employers adapt their interview practices.
Key findings
The feasibility study with 9 autistic and 8 neurotypical participants revealed several important findings. The Conversation Management System successfully understood 93% of interviewee responses, and the physiological stress detection model achieved 84.5% accuracy. However, autistic participants reported significantly lower perceived usability (SUS score 57.5 vs 77.5, p=.011) and lower comfort and confidence during interviews compared to neurotypical peers. Notably, the intentional interruption feature—where a receptionist knocks to announce a phone call—was reported as frustrating by autistic participants, validating its purpose as a realistic stressor for practice. Participants valued the structured interview flow and whiteboard-based problem-solving tasks. Some noted the avatar's voice sounded "robotic," suggesting improvements needed for naturalism. The participatory design process was extensive: 23 stakeholders (autistic individuals, job coaches, employers) informed conceptualization, 4 autistic self-advocate interns provided feedback during development, and the feasibility testing further refined the system. This co-design approach surfaced specific needs like the ability to request question repetition and mechanisms for expressing oneself nonverbally.
Relevance
CIRVR represents an important shift in employment technology for autistic individuals: rather than solely training candidates to mask their differences, it also aims to provide employers with insight into the interview experience from an autistic perspective. The Dashboard's visualizations of stress levels, gaze patterns, and emotional responses during specific questions could help employers identify which interview practices cause unnecessary difficulty without revealing relevant skills. For practitioners, this research highlights the value of physiological sensing as an objective measure of interview stress—important because autistic individuals may not show stress through typical facial expressions or body language. The significantly lower usability scores from autistic versus neurotypical participants also underscore the need to test accessibility technologies specifically with the target population, not just neurotypical proxies. Current limitations include the lack of efficacy data (the system works, but does it improve real interview outcomes?) and the stress model being trained on neurotypical data. The researchers acknowledge these gaps and plan longitudinal studies at vocational agencies. The bidirectional design—serving both job seekers and employers—offers a model for employment accessibility interventions that address systemic barriers, not just individual deficits.
Tags: autism spectrum disorder · virtual reality · employment · job interviews · stress detection · eye tracking · affective computing · speech recognition