AI Clinical Assistant

Built#ai

Anthropic client (claude-opus-4-8, mock↔live) suggests ICD/HCC codes from a note INTO the staging gate — a human still promotes. Plus an eCQM care-gap panel. ai-service behind ai:suggest/ai:read.

Standard of work · ISO/IEC 25010 — Meets 9/9
  • 1. Self-identifyingUsability — appropriateness recognizabilityThe page names itself: breadcrumb, title, section, status badge, and a page id.
  • 2. Navigable forward & backUsability — operabilityBack/Next step through the product; breadcrumb + active sidebar; browser back works.
  • 3. Demo-prefilled — no empty statesUsability — learnabilityEvery form is pre-filled with realistic demo data so the screen is never blank.
  • 4. Accessible — AAA, large fonts, big targetsUsability — accessibilityLight mode, 7:1+ contrast, ≥44px controls — built for a 76-year-old physician.
  • 5. Clear result feedbackFunctional suitability — appropriateness; Reliability — fault toleranceEvery action maps a Result to a visible success/error message; codes → HTTP status.
  • 6. Secure by default + sign-outSecurity — confidentiality & authenticityDefault-deny permissions, practice-scoped data, a sign-out control, no PHI in logs.
  • 7. ResponsivePortability / Compatibility — adaptabilityWorks on desktop, tablet, and phone; the sidebar collapses on narrow screens.
  • 8. ResilientReliability — availability & recoverability; Performance efficiencyLoading and upstream-down states are handled — never a blank or frozen screen.
  • 9. ConsistentMaintainability — modularity & reusabilityOne shell, one set of design tokens, shared components across every page.
AI suggests, a human promotes. Suggestions are never added to the chart automatically — the accepted codes are staged into the Intake gate for review, exactly like an imported document.

Clinical note

Paste or dictate the encounter note. The AI extracts supported ICD-10 / HCC codes — it does not diagnose.

AI suggestions

Analyze a note to see suggested codes here.