HKUST unveils ASD Super Agent, an AI platform for autism screening and support

Researchers at the Hong Kong University of Science and Technology have developed “ASD Super Agent,” which the university describes as Asia's first integrated AI platform for autism support, Mirage News reported on November 11. The system combines early screening, personalized intervention and continuous family follow-up in one place.

The project is led by Prof. Kani Chen of the Department of Mathematics and the Department of Industrial Engineering and Decision Analytics. During development the team visited special education institutions, child development centers and psychiatric hospitals across Hong Kong, the Greater Bay Area and Beijing, and spoke with nearly 500 parents of children with autism.

Those conversations surfaced a familiar set of problems: long waits for assessment, high treatment costs and fragmented information about intervention services. Autism spectrum disorder affects roughly one in every 100 children worldwide, and delays can mean missing the most effective window for support.

Screening is designed to happen at home. Parents record short videos of a child's daily activities on a smartphone or smart device, complete standardized questionnaires and provide basic developmental information; the platform then generates a risk assessment and analytical report they can take to a professional evaluation. In preliminary testing the system reached an AUC of 0.92, and the team says further clinical studies and real-world validation are needed before deployment.

Intervention recommendations rest on a “20+1” framework. The 20 refers to more than 20 internationally recognized autism intervention models embedded in the system, including Applied Behaviour Analysis, Naturalistic Developmental Behavioural Interventions, TEACCH structured teaching and a localized BSR model. The +1 lets therapists, special education providers and organizations add their own proven methods to the knowledge base.

Each recommendation can come with relevant academic literature and background references, so families can hold better-informed conversations with clinicians and educators. The platform also builds a dynamic developmental profile from text records, video assessments and developmental information, breaking long-term goals into daily actions and tracking progress over time.

Privacy was a design constraint: the platform follows a data-minimization approach, prioritizing on-device posture feature extraction and processing only the information needed for analysis. Prof. Chen said the aim is to give parents practical, evidence-based support rather than replace professional care.

Using AI for developmental screening carries obvious promise and obvious limits. It can shorten the path to information, but a risk score is not a diagnosis, and an AUC of 0.92 from preliminary testing is not clinical proof. The questions to watch are the scale and results of further validation, when the platform actually deploys, and how HKUST handles medical compliance, data protection and cross-cultural applicability.