AI Medical Tourism Platform for the United States

Client

HealthVoyage

Scope

AI/MLMulti-tenant SaaSHealthTechConversational UX
Checkup Trip
Chat

Ask Checkup Trip…
01The situation

Finding affordable, vetted medical care in the US is unreasonably opaque. Pricing is hidden behind insurance contracts, clinic quality is hard to compare, and patients shopping for elective procedures (implants, LASIK, cosmetic surgery, hair restoration) are forced to navigate a fragmented network of provider directories, review sites, and unanswered call-backs.

Form-led directory tools don't solve it: filters are coarse, prices are vague, and the user always has to know what they're looking for before they ask. The medium itself fights the user.

Insight. This isn't a search problem. It's a conversation problem.

02What we built

A conversational medical-tourism platform with three connected surfaces and a verified clinic registry

We replaced the directory model with a chat-first experience patients can speak to in plain English. The AI translates intent into a structured query, runs it against a normalised procedure index, and returns vetted clinics with transparent pricing.

Patient surface. A single chat interface with shareable shortlists, save-to-account, and one-tap callback requests. No commission fees, no middlemen.

Clinic portal. A self-serve workspace where vetted clinics manage their profile, procedures, and inbound leads. Status tracking, CSV export, and a side-drawer detail view per inquiry.

Admin dashboard. Conversion funnel from chat-open to lead-submitted, plus top-procedure / top-location panels and feedback ratings. The whole platform's health on one screen.

AI infrastructure. A FastAPI + Ollama (llama3) intent service in front of the database. Natural-language queries get classified, matched, and re-framed, so patients get a useful answer, not a database dump.

03Surface tour

Patient, clinic, admin: one shared data model.

Three portals, one Prisma schema. A clinic update propagates instantly to chat results; a patient lead lands in the clinic inbox in real time. Switch tabs below to step through each surface, or use ← / → keys.

For patients

Plain English in. Verified shortlist out.

No forms. The AI translates intent into a structured query and only returns vetted clinics.

Dental implants in NYC under $4k

Found 3 verified clinics · top match 4.9 ★

Manhattan Smile Studio

Manhattan, NY · ★ 4.9

$2,400

Liberty Dental Group

Midtown, NY · ★ 4.8

$3,150

Coverage

500+

vetted clinics across all 50 states

Specialties

12+

dental · LASIK · cosmetic · hair · more

Availability

24/7

AI assistance, no business hours

04How we got there
1

Started from the conversation, not the directory

Form-led search assumes the user knows what to type. We built around the moment a patient says 'I think I need this, in this city, for around this much', and let the AI take it from there. The chat surface came first; the database structure came after.

2

Designed three portals around one shared data model

Patient, clinic, admin all read and write the same Prisma schema. One set of leads, one source of truth. The portals are different lenses on the same data, not parallel apps. A clinic update propagates instantly to the chat results; a patient lead lands in the clinic inbox in real time.

3

Treated AI as plumbing, not magic

The intent layer (FastAPI + Ollama) does one job well: classify the patient's request and structure it. Search runs on the database. Result framing (disclaimers, follow-up questions, fallback suggestions) uses the LLM again. Each piece is replaceable; nothing is mysterious.

4

Built clinic vetting into the workflow

Every clinic on the platform passes a credential review before going live. Status flows (pending → approved → suspended) are first-class in the schema, not a content moderation afterthought. The 'Verified' badge in the chat results is enforced at the data layer.

05Under the hood

From a typed question to a verified shortlist, in five steps.

The chat surface is the front door. Behind it, a pipeline splits intent from search and re-frames the result so the patient sees a useful, legible answer, not a database dump.

Step 1

Patient query

Natural language. Voice or typing.

Step 2

Intent resolver

FastAPI · Ollama (llama3) · classification

Step 3

Search & match

Prisma · MySQL · normalised procedure index

Step 4

AI compose

Result framing · disclaimer · follow-up

Step 5

Verified results

Clinic cards · pricing · save & request

StackNext.js 16TypeScriptFastAPIOllama (llama3)PrismaMySQL 8DockerNginx

Let's talk about what's next

Tell us where you're headed and what's in the way. You'll get a concrete take, not a sales pitch.