CareLink — Responsible AI

Responsible AI guardrails

CareLink never sends raw survivor intake to AI.

CareLink separates sensitive intake from planning support. AI only works from a human-approved, minimum-necessary case abstraction built with synthetic demo data. Every recommendation is a draft until a survivor and social worker decide what moves forward.

Privacy boundary

No raw intake to AI: names, addresses, narrative details, case notes, and other direct identifiers remain in the human-controlled intake layer.
Local/mock privacy filtering: this demo uses a local browser-side filter simulation to show what is removed before planning.
Human-approved abstraction: a social worker reviews the filtered summary and approves only the minimum facts needed for goal planning.
Synthetic data only: demo scenario details are fictional and contain no real PHI or PII.

Decision boundary

AI output is draft only: pathway suggestions are planning prompts, not final service decisions.
Survivor/social worker final decision: goals must be reviewed and approved by the survivor and social worker.
Consent-based sharing: agencies only see consented, goal-specific information needed for their task.
Progress transparency: survivors can review goal status and see what was shared.

Architecture diagram

Explainable privacy-preserving pathway

Raw Intake
Human-controlled

→

Privacy Filter
Local/mock

→

Human Review
Approve minimum facts

→

Case Abstraction
No raw narrative

→

AI Pathway Draft
Draft only

→

Survivor/Social Worker Approval
Final decision

→

Goal-Based Agency Channels
Consent-limited

→

Progress Feedback
Visible to survivor

Survivor view

Plain-language cards show approved goals, next steps, consent, and progress without operational case jargon.

Social worker view

Operational review tools show what is blocked, what needs approval, and which goals are ready for coordination.

Agency view

Limited task cards show only the consented goal-specific request, due date, and status update controls.