A behaviour-adaptive AI assistant enhancing accessibility and usability for blind users through real-time interaction personalization

Individuals with visual disabilities are still struggling when using the modern AI-driven system as most current assistive technologies are based on a static paradigm of interaction that fails to adjust to the behaviour of individual users. Recent developments on large language models (LLMs) present even more chances in the sphere of accessibility, though their combination with adaptive, behaviour-sensitive interaction is under-researched. This study presents AURA (Adaptive User-Responsive Assistant), an accessibility system based on voice that integrates an interaction based on LLM with personalization based on real-time behaviour. The speech rate, verbosity and language complexity are dynamically modified by AURA in response to behavioural cues through replay actions, skips, and length of time listening, allowing within-session adaptation. The proposed approach was evaluated through an exploratory, simulation-based study using representative behavioural profiles to examine the system’s adaptation capability under controlled conditions. Quantitative results indicate that behaviour-aware personalization can reduce replay and skip events, improve interaction efficiency, and reliably converge toward intended interaction profiles when compared to a non-adaptive baseline. These results are made as initial, evidence-of-concept, which indicates the possibility of integrating LLM-based interaction and behaviour-sensitive adaptation in accessibility-focused systems. While the current study does not involve human participants, it establishes a conceptual and technical foundation for future research. Ongoing and future work will focus on human-subject evaluation with blind and visually impaired users, in-the-wild deployment, and extension to more complex conversational and multilingual accessibility scenarios.