Call centres and mental health chatbots are about to get a lot better at reading the room. The CRYSTAL project is building a new generation of conversational AI that can do more than answer simple questions—it will detect a user’s emotional state and respond with empathy, whether that user is an anxious older adult seeking emotional support or a frustrated customer on a helpline. Current chatbots are brittle: they fail when topics shift, and they cannot sense frustration, sadness, or confusion. CRYSTAL closes this gap by combining large language models with user state estimation—tracking tone, hesitation, and behavioural cues—and then using that information to steer the conversation. The system will learn to adapt its language and tone on the fly, without needing to be retrained for every new scenario. If successful, the technology could transform mental health triage, ageing-in-place support, and customer service operations. Call centres could handle complex, emotionally charged calls more accurately and humanely. Vulnerable individuals could access reliable, non-judgemental conversational support at scale. The project also prioritises ethical safeguards: all conversational data will be anonymised, encrypted, and audited for bias and robustness.
View original technical description
CRYSTAL (Conversational Systems for Emotional Support and Customer Assistance) will advance research for a next generation of conversational models and systems that implement reliable, friendly and efficient human-machine interactions that can assist humans in tasks beyond automatic information provision, such as emotional support as well as customer assistance. Emotional support provided to vulnerable individuals, such as ageing adults or people experiencing anxiety or depression. Advanced customer assistance provided to call centres, providing customer care and response to customer inquiries, in an empathic, inviting, accurate and efficient manner. Conversational models and systems will be developed by proposing novel learning strategies for domain adaptation, exploring Large Language Models (LLMs) and prompt engineering methodologies. User state estimation will be also considered, and strategies, such as behavioural models and stakeholder goals, will guide the policies for dialogue management. Novel evaluation frameworks will be also designed including automatic metrics, usability as well as human acceptance of the technology. Strict ethical and AI robustness compliance procedures will be applied, as conversational recordings will be collected, anonymised, annotated and encrypted for collaborative work among the project partners. The project will apply extensive staff secondments and foster sustainable knowledge exchange. It will grow the experiences and insights of the partner organisations’ staff via intersectoral and interdisciplinary collaborations, implemented through secondments as well as training and networking activities. Dissemination, communication and exploitation activities will expand the project’s impact pathways.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know