Web App

Pathify

A clinic appointment and doctor-matching platform from a symptom-based AI checker

Details

Role

Project Lead & Frontend Developer

Project Type

Hackathon Project

Duration

24 Hours

Status

MVP / Completed Prototype

Recognition

AI.DEAS for Impact 2025 Hackathon, Region VI - Regional Finalist

Built With

JavaScript

Tailwind

HTML

CSS

Python

Flask

MySQL

Tensorflow

PyTorch

Scikit-learn

What it is

A 24-hour hackathon prototype that helps patients figure out where to go and find a specialist. AI-assisted screening suggests possible conditions, then maps them to a specialty. Patients who already know what specialty they need can skip screening.

From there, TOPSIS ranks the top 5–10 doctors using the patient's preferences, and the physician can accept or reject the referral. The screening is guidance, not a diagnosis or replacement for a physician.

The problem

Context

Finding a specialist often means word of mouth, wide directories, phone calls, and whatever the patient or family prefers. People walk into a general visit without a sense of what they are feeling, then still have to hunt afterward.

Impact

Finding a doctor takes longer than it should, unanswered requests stall, and a rejected or ignored referral has no automatic next step.

What I did

I led the frontend and connected the patient and physician sides into one referral flow, from screening to a request the physician can answer.

Planning the patient path from screening or skip, through preferences and ranked doctors, to a referral request.
Building the patient and physician interfaces.
Connecting symptom screening and specialist matching to the referral flow.

Coordinating frontend work with the team's model and backend services.
Tying doctor ranking, referral requests, and physician accept or reject actions into one flow.

How I designed it

Allow a skip if the patient already knows

Screening is a Yes/No questionnaire. Patients who already know the condition or specialty can go straight to preferences.

Split condition prediction from specialist matching

One model suggests possible conditions as a prediagnosis, not a diagnosis. Another maps those conditions to a specialty so the patient knows where to go next.

Show a short ranked list, not a directory

TOPSIS uses location, gender, experience, schedule, insurance, and teleconsult options, then returns the top 5–10 doctors.

Keep the physician in the loop

Accept, reject, or ignore for 24 hours. A reject or timeout starts another referral instead of leaving the patient stuck.

Main features

Symptom screening

Yes/No questions feed a model that suggests possible conditions as a prediagnosis. Guidance for where to go, not a replacement for a doctor.

Doctor ranking

TOPSIS returns the top 5–10 doctors using location, gender, experience, schedule, insurance, and teleconsult.

Physician portal

Physicians verify, manage schedules, and accept or reject incoming referral requests.

Automatic re-referral

A rejected request or a 24-hour timeout starts another ranked doctor instead of stalling.

Preference filters

Patients set specialization, location, gender, experience, schedule, insurance, and teleconsult before ranking.

Built with

JavaScript and Tailwind

Built the patient and physician interfaces, screening questions, preference filters, ranked doctor lists, and referral interactions.

Python and Flask

Handled the APIs for symptom analysis and condition-to-specialist matching.

Tensorflow and PyTorch

Powered the team-trained symptom-to-condition model used during screening.

MySQL

Stored referral, specialist, schedule, and appointment data.

Scikit-learn

Supported the model pipeline alongside TensorFlow and PyTorch.

Kaggle & Hugging Face datasets provided public datasets used by the team to train the screening models.

Hard parts and how I solved them

Fitting a clinic product into 24 hours

A full records system was out of reach, so the build stayed on one working referral path that a judge could follow.

Matching more than a specialty name

After a specialty is chosen, TOPSIS ranks doctors on the patient's own filters so the shortlist can be explained.

Keeping patient and physician on the same referral

Accept, reject, and timeout all write back to one request so the next doctor is not a disconnected search.

Combining a prediagnosis with a doctor finder

Screening had to stay guidance, not a diagnosis, while still pointing the patient at a ranked specialist instead of a directory.

How it works

1
A patient starts screening or skips it
Yes/No questions collect symptoms, or the patient goes straight to doctor preferences.
2
Conditions map to a specialty
A model suggests possible conditions, then a second step picks the matching specialization.
3
Doctors are ranked
Preferences plus TOPSIS produce a top 5–10 list instead of a full directory.
4
A referral is sent
The chosen doctor gets the request. The patient can track whether it is waiting, accepted, or moved on.
5
The physician answers or time runs out
Accept, reject, or no reply in 24 hours. Reject and timeout start another ranked referral.

What came from it

The hackathon produced a working prototype that gives a prediagnosis, a short ranked list of specialists, and a referral the physician can answer.

It was selected as a Regional Finalist in AI.DEAS for Impact 2025, Region VI, and later led to the AI Triage. It remains a 24-hour MVP, not a deployed clinic system.

What I learned

A short build works better as one path. Screening to referral is clearer when the rest of a clinic product is left out.
Doctors sit in scattered clinics. Referrals still run on word of mouth, wide directories, and calls, so a short ranked list is easier to act on.
Doctor registration still needed a place. Admin and verification were taken into account even without building a full admin product.

Screenshots

Home

Displays the main landing interface for patients and doctors, including symptom input and navigation to referral tracking.

Next project

Investracture

A space for entrepreneurs and investors to discover opportunities.