Designing responsible AI for high-stakes child welfare decisions.

Child welfare intake happens in real time. A worker may be listening to a caller report homelessness, abuse or another concern while simultaneously capturing names, addresses, issues, context and required documentation.

The opportunity for AI was not to automate the decision. It was to help the worker capture and organize information, surface useful prompts and recommendations, and prepare documentation while preserving review, override and professional judgment.

The experience begins with the work already happening: the intake call. Live transcription creates a working layer where AI can identify potentially useful information without writing directly into the official form.

Recommendations and extracted information need to be understandable at a glance. Flags and color-coded states help workers identify scenarios quickly, while the interaction model makes review and override explicit.

The workflow extends beyond transcription. AI can help prepare documentation, translate reviewed information into narrative, identify matched referrals for the same child, surface recurring patterns and support closing the referral loop.

I lead and mentor five designers across Closed Loop Referral, Enhanced ChildLine, Mandated Reporter, API/Data Services and SNAP initiatives. My role spans product direction, executive workshops, design quality and alignment across policy, accessibility, technology and business priorities.

The work positions Pennsylvania as an early model for responsible AI adoption in government while creating a practical experience framework for how AI enters caseworker practice: visible assistance, explicit review, human override and accountability at the point of decision.

Visuals are reconstructed and anonymized to protect client confidentiality.