DSH / AI × HEALTHCARE

Practical care navigation.
Private, by design.

CareRoute helps people in the US find affordable-care resources and organize questions for a visit. A small on-device model suggests administrative tools; a separate warning checklist takes priority.

Three-minute judge walkthrough

  1. Type “I do not have insurance and need help with costs.” Apply the affordable-care suggestion. Build a plan and inspect the HRSA and 211 links.
  2. Switch to Prepare for a visit. Add “What should I bring?” and “Who can explain the fees?” Build the card. Your own questions appear first, in order. Print it.
  3. Add and remove questions. A maximum of five encourages prioritization. Editing invalidates the previous card; reset clears all drafts.
  4. Select both a physical warning and a crisis warning. The 911 action appears immediately and stays primary. 988 appears as additional support.
  5. Clear the draft or reload. Your concern and questions disappear. No account, database or external model API is involved.

What the AI actually does

Multinomial naive Bayes learns token counts from 24 synthetic administrative phrases. It suggests affordable-care or visit-preparation tools and abstains on weak or unknown matches. It does not predict diagnoses, classify symptoms, or determine urgency. Users choose whether to apply a suggestion.

Healthcare value and limits

The prototype connects resource discovery with an actionable question card. It does not verify clinic availability, fees or medical appropriateness. Resources are US-focused. No patient outcomes, clinical validation, adoption or revenue are claimed. The next step would be usability testing and professional review.

October 7 additions

The new question organizer supports adding, removing, limiting and prioritizing up to five questions. It includes drafts in reset behavior and invalidates stale cards. The public GitHub source and this DSH walkthrough make the implementation inspectable.

Origin and disclosure

The interface existed before this entry; the October 6 release added the local model, warning fixes and Vercel hosting. This is an extension of that project, which was already submitted to UnivaBio, ML Empowerment and Next Founders. Codex assisted with implementation, testing and writing. Dependencies are listed in package.json.

The October 6 demo shows the core tools; the October 7 question organizer is shown in the live app and release screenshots.