WiseCareAI

WiseCareAI is a health-insurance platform for the U.S. market. My work was end-to-end product design: research, IA, flows, interaction design, and a shared design system across agent tools, internal ops, and the public site. Generative UI/AI was integrated to augment core workflows—not define them.

Product scope:

ACA Enrollment (Agent): guided quoting/enrollment with dynamic form blocks. Medicare Quoting (Phone): quick compare with rationale for seniors. Internal Ops (CRM): timeline/status insights from notes and events. Marketing Site: clear story and trust signals in a regulated domain. Design System: tokens, components, patterns (including generative states).

Responsibilities:

User interviews and task flows (agents, support, end clients). Information architecture and navigation across tools. Wireframes → hi-fi UI → prototypes; dev handoff. Design system definition and maintenance. AI feature integration: prompts/guardrails/UX for explainability and edits.

Where AI/Generative fits (as augmentation):

Adaptive steps/fields when case context changes. Free-form intent or guided inputs—both converge to the same editable draft. Rationale overlays: why a plan is suggested (cost, coverage fit, eligibility). Recovery loops: quick fixes to model mistakes without losing progress.

Key design challenges:

Stability vs plasticity: keep anchors while the work area adapts. Trust without overload: show a one-glance rationale; expand for trace. Preview vs commit: drafts first; explicit, reversible writes. Policy complexity: ACA/Medicare rules without exposing regulation-speak.

Pattern: Anchor Layout — persistent navigation/header/progress; only the canvas mutates. Reduces disorientation while supporting flexible steps.

Pattern: Preview → Refine → Commit — AI creates drafts (plans/forms/messages). Agents adjust inline chips (filters/constraints) before any system write.

Explainability surface: compact rationale (cost, fit, eligibility) with tap-to-expand trace to sources and constraints.

Interaction modes: free-form need statements or guided steps. Users can switch modes any time; both routes land on the same draft.

System behavior (with engineering): prompt scaffolds, eligibility guardrails, deterministic anchors, streaming feedback, error classes mapped to recovery UI.

Patterns distilled:

Correction Loop — inline edits update the draft and current context. Anchor Layout — stable shell; generative canvas. Preview-First — drafts by default; explicit commit. Rationale Card — concise “why”, expandable trace. Low↔High Agency Toggle — guided steps or free-form intent. Memory Notes — lightweight notes the agent and system can reference.

Impact (early signals):

≈25% faster enrollment time in agent tests. Fewer back-and-forths due to visible rationale and editable constraints. Lower recovery cost by treating AI output as drafts, not decisions.

Marketing: straightforward, credible narrative; visuals aligned with product explainability and regulation sensitivity.

Outcome: a coherent multi-tool product with a design system that embeds generative capabilities where they help most, while keeping control, speed, and trust.