JMIR Publications

ABSTRACT

Background:

As more autistic youth age out of school-based supports, families must navigate fragmented health, education, employment, housing, and benefits systems. This burden falls especially hard on families facing language barriers.

Objective:

We developed StarMate星伴, a publicly deployed custom GPT in ChatGPT, to serve as a multilingual transition-navigation assistant for autistic youth and caregivers.

Methods:

The initial build uses uploaded knowledge files and retrieval-backed response generation to ground answers in curated material rather than relying only on model memory. Local knowledge assets include English guidance on employment-related self-advocacy and a Traditional Chinese resource guide organized around New York transition services such as The Arc New York, OPWDD Front Door, and ACCES-VR. The tool's public prompt starters focus on multilingual support in New York City, the end of school-based services, housing planning, and OPWDD application support. We conducted a formative appraisal of deployment metadata, local knowledge assets, and workflow design.

Results:

This appraisal confirmed that StarMate is configured as a tailored generative AI assistant for transition-to-adulthood navigation and fits the broader category of retrieval-augmented, knowledge-grounded assistants.

Conclusions:

StarMate's main strengths are low access barriers, multilingual reach, and a practical focus on action steps and agency navigation. Its limitations are equally important: it does not replace clinicians, case managers, or legal advisors; resource freshness must be maintained; and formal human evaluation of accuracy, safety, and usefulness remains necessary. StarMate illustrates a feasible public health tool pattern for culturally responsive transition support while defining a clear agenda for rigorous next-phase evaluation.