06 / Software
Healthcare navigation software
I built a patient-facing clinic-search assistant with conversational preferences and application-controlled retrieval and ranking.
Patient-navigation workflow
- 01 / Request
Describe the care needed
Service request and location or university; refine language, proximity or affordability.
- 02 / Interpret
Update intent and preferences
Carry relevant context forward, while giving a new non-search question its own response route.
Choose the response path
Clinic search
- Retrieve clinic listings
- Filter and rank through application logic
- Show options, explanations and data gaps
Coverage guidance
- Explain what to check with the provider or insurer
- Cannot see a policy or confirm eligibility
Booking guidance
- Hand off to clinic profiles and contact details
- Does not complete appointments
Additional routes: FAQ responses and bounded, non-diagnostic urgent guidance.
These diagrams are simplified reconstructions of my patient-navigation work.
01 / The brief
Turn a care request into search preferences.
I used server-side extraction to turn conversational requests into structured preferences, then kept clinic retrieval and ranking in application logic. Users could refine location, language and affordability cues, request closer options, or ask for more listings.
This patient-navigation work is separate from Careviv’s current doctor–clinic matching product.
02 / My contribution
Bound the work behind each response.
I gave new non-search questions precedence over the previous search context, limited external calls and search passes, and used rule-based interpretation when extraction was unavailable. Supabase supplied the clinic listings.
I also adapted existing web/backend workflows into mobile-facing onboarding, family-doctor access, waitlist and clinic-discovery features.
Implementation details
Receive the request
A Next.js assistant API receives the message and available conversation context.
Interpret within a bounded call
Server-side structured extraction uses a one-call guard and timeout. Rule-based interpretation provides a fallback when extraction is unavailable.
Route, retrieve and rank
Application logic selects a search or guidance route. Supabase supplies clinic listings; filtering and ranking remain deterministic. Search passes and fan-out are capped.
Present options and next steps
Return clinic cards and explanations, or a response appropriate to the guidance route.
03 / Evidence & scope
Refine the search, then inspect the result.
I tested language, proximity, cheaper-first and more-options requests. Sparse language information required widening the search and explaining gaps. Some preference refinements returned the same ordering, which showed where the ranking needed further work.
Next project / 07