B Brie FDE Cases 한국어
Customer Case Portfolio · Healthcare

Eye clinic inquiries,
answered first by AI, 24/7

We automated KakaoTalk customer inquiries for Dongtan First Eye Clinic. AI answers questions about exams, doctor schedules, surgery prep, costs, and documents instantly and accurately, and automatically routes sensitive inquiries — appointment confirmation, symptoms, emergencies — to the front desk.

🏢 Dongtan First Eye Clinic 👥 Patients · guardians · front-desk staff 🧪 Built and validated (preparing activation)
98.3%Inquiry-type classification accuracy
0Medical safety rule violations
~7.9 secAverage response time
−78%AI operating cost saved

※ The figures above are from pre-launch performance evaluation (scenario / golden set).

The Challenge

The repetitive inquiries the front desk carried

The same questions repeated all day, and after-hours inquiries had to wait until the next day.

1

The same questions all day long

Front-desk staff had to answer, by hand, every question about doctors' weekly schedules, pre- and post-surgery precautions, cost ranges, and issuing medical certificates and opinions.

2

No response after hours

Inquiries that came in at night or on holidays sat until the next business day. Patients had to wait for an answer.

3

A wrong answer to a medical inquiry is an incident

A wrong phone number or off-the-cuff medical advice reaching a patient isn't a simple mistake — it's an incident. The burden of always being careful fell on people.

The Solution

Answer what it can instantly, hand off what it should

AI responds instantly, 24/7, to inquiries it can answer, and automatically passes inquiries that need human judgment to the front desk.

STEP 1

Kakao inquiry received

The moment a patient asks on KakaoTalk, it's received.

STEP 2

Classify · safety check

Judges the inquiry type and verifies medical safety.

STEP 3

AI response / desk judgment

Answers if it can; hands off if sensitive.

STEP 4

Send / queue

Send the answer, or add to the front-desk queue.

Feature ① · Core

24/7 instant response + human-handoff judgment

When a patient asks on Kakao, a "checking now" notice appears immediately and the AI sends an answer. It only auto-answers inquiries it can handle, and automatically routes sensitive ones — appointment confirmation, symptoms, emergencies, refunds — to the front desk.

  • Instant responses at night and on holidays, with no waiting
  • The desk focuses only on inquiries that truly need a person
  • No inquiry is dropped — safely handed to the queue
Patient KakaoTalk question24/7 · including proxy inquiries
↓
Classify + safety checkAnswerable? · sensitive?
↓
AI instant answerSensitive inquiries auto-routed to the desk
Answer what it can instantly, hand off what it should
Feature ②

Doctor schedule guidance

For questions like "which days does Dr. ○○ see patients?", it reads each doctor's weekday, AM/PM clinic, and surgery schedule and answers precisely.

  • No more front-desk staff repeating the doctor × weekday schedule
  • Guides available days by visit purpose (exam / surgery)
  • Fewer wasted trips and rebookings from day mix-ups
Per-doctor weekday clinic / surgery schedule table (doctor names masked)
The doctor schedule the AI references (doctor names masked)
Feature ③ · Safety

Answers grounded in the clinic's official materials

The AI doesn't make things up — it finds grounding in the surgery guides and precautions the clinic actually uses. Every answer is verified once more before sending, and phone numbers are always corrected to the clinic's main line.

  • Matches the clinic's official wording — never invents information
  • Guides costs as ranges and routes precise quotes to consultation
  • Re-verifies before sending → on failure, routes to the desk with a safe message
  • Zero medical safety rule violations in performance evaluation
Surgery prep / recovery questionLASEK · retina · oculoplasty, etc.
↓
Search official clinic materialsGround the answer in actual guide documents
↓
Grounded answer + re-verifyAbnormal symptoms routed to the desk
Doesn't invent — answers from the clinic's documents
"A wrong answer reaching a patient isn't a simple mistake — it's an incident."

So the AI only answers what it can, cross-checks its grounding, and hands off to a person when it isn't sure.

Seen in data

The AI answers from the clinic's official materials

These are the actual clinic materials that ground its answers. Because the AI finds its content in these documents, it never diverges from the clinic's own wording.

Pre-retina-surgery precautions guide
Pre-retina-surgery precautions
Oculoplasty pre- and post-surgery guidance
Oculoplasty pre- / post-surgery guidance
Post-LASEK precautions and emergency response guide
Post-LASEK precautions · emergency response
The Impact

No waiting — but safe

AI clears away repetitive inquiries, and when human judgment is needed, it always goes to a person. (Figures below are from pre-launch performance evaluation.)

BEFORE
  • Responses only during business hours — after-hours inquiries wait
  • The desk answers repetitive inquiries by hand every time
  • Guidance accuracy varies by staff member
AFTER
  • 24/7 instant responses (avg. ~7.9 sec)
  • 7 major inquiry types auto-answered (95% auto-resolution)
  • 98.3% inquiry classification · 0 medical safety rule violations
Medical safety
0

In performance evaluation, there were no cases of wrong medical advice, false information, or wrong numbers reaching a patient. It's designed to hand off to a person, not answer, when it isn't sure.

Status

Built and validated, preparing activation

The consultation pipeline and safeguards are validated; we're preparing full activation through Kakao channel integration, loading real guide data, and a medical-law review.

Tech Stack

Built on a proven stack

The keys are search that finds grounding in the clinic's materials, and multiple safeguards that block wrong answers.

Python · Django PostgreSQL + pgvector (RAG) Claude Sonnet · Haiku OpenAI embeddings Kakao i Open Builder Celery · Redis LangSmith standarda-core