GabAI

Inspiration

Our project is based off our personal experience in witnessing first hand the effects of an overextended healthcare system. The Philippine public healthcare system is very bureaucratic, you would need document/s x from department y to get checked, and you would only find out after you already spent hours waiting in line. During our volunteer service in the PGH, we've had multiple patients who are confused and distraught because they only just figured out they lacked certain documents necessary to be treated, wasting an entire and days worth of travel expenses. Based on the real experiences of our patients, we built this project in hopes that they would not spend hours travelling and waiting in line for nothing ever again.

What it does

GabAI is a medical companion web app that acts as a companion for low-to-middle-income Filipinos through every single healthcare encounter; before you leave, while you're there, and after you leave.

Before leaving, GabAI will give you a list of facilities you can go to with directions and estimated costs, and required documents to follow protocol in availing the healthcare they need.

While they're at the clinic/hospital, the app can remind the patient of their rights and/or benefits that they can avail depending on the institution they are in and nature of the visit. The app can also transcribe and summarize the patient doctor discussion and return a simple to understand "To Remember" cards for prescriptions, referrals, and lifestyle changes.

After the appointment, the app would give a simple cogent summary for not only the patient, but also for the next physician that would handle the same patient. It would give reminders to take medicine at the prescribed times, ask about your symptom progression. Should you be referred by your previous physician to go to a different institution with a certain specialist, the app would take into account everything that happened in your most recent appointment and start over from the "before" phase again.

How we built it

We built GabAI using a modern, performance-oriented stack to ensure it remains accessible even on budget mobile devices. The core of the application is powered by Next.js 15 using the Turbopack bundler for near-instant development and optimized production builds. For the intelligence layer, we orchestrated multiple agents using Gemini 2.0 Flash, which handles everything from classifying vague symptom descriptions to interpreting photos of handwritten prescriptions through its multimodal vision capabilities.

To solve the complex problem of facility routing in Metro Manila, we implemented a custom spatial engine that calculates the distance between the user and our database of 47 verified healthcare facilities. We utilized the Haversine formula to determine the precise distance $d$ between coordinates, where $d = R \cdot 2 \cdot \operatorname{atan2}(\sqrt{a}, \sqrt{1-a})$ and $a = \sin^2(\frac{\Delta\phi}{2}) + \cos\phi_1 \cdot \cos\phi_2 \cdot \sin^2(\frac{\Delta\lambda}{2})$. This allows us to provide a relevance score that balances physical proximity with the specific services a patient needs.

The "Angkla persistent memory feature was implemented using a combination of Zustand for client-side state management and Firebase Firestore for long-term storage. By maintaining a structured history of every encounter, the app can prep patients for their next visit by injecting past diagnoses and medications into new consultation scripts. We also integrated the Web Speech API to enable a voice-first experience, allowing users to speak naturally in Taglish while the AI provides high-quality speech synthesis in a warm, local tone.

Challenges we ran into

Our first problem was about patient data privacy, as one of our core feature needs to listen in on the discussion between patient and doctor in order to return a summarized transcription for the patient to use. We also had to account for response validity, even though we took special precaution to ensure we don't do any diagnosis, we had to make sure that some of the suggestions and summarizations of GabAI is strictly context and physician based. We also lacked data to give actual travel expenses of a user to a specific hospital, so we had to simulate it using OpenStreet and custom equations to simulate the distance aligning with patient needs.

Accomplishments that we're proud of

Most health apps fall into two categories. First, information portals—they dump medical knowledge on users and assume high health literacy. Second, appointment booking apps—they assume the healthcare system is organized and integrated, which it isn't in the Philippines. We do neither. We're a companion that assumes the user is confused. Here's the accomplishments that we're proud to say are unique to our project: (1) Voice-first interaction— the core accessibility layer, you don't have to take notes anymore, just press and record and it will return more high yield information with less effort on the patient's part. (2) Persistent memory—every visit is remembered, each appointment is built off the previous one, so the next appointment doesn't start from zero. (3) Free and anonymous—no signup wall, no data harvesting, secure data. (4) Designed for the fragmented system—we don't assume the Philippine healthcare system is organized, we work with the idea that it's terribly stretched thin just like the PGH. Other apps try to fix the system or info dump on the patient. We help patients navigate it as it exists today.

What we learned

Patient Data Privacy is Solvable — Patients want recordings if they own the data and trust the system Fragmentation is Systemic — Even within hospitals, every handoff is a restart Patients Want Instructions, Not Information — "What do I do?" beats "What is this?" Healthcare Workers are Allies — They want patients prepared; it makes their jobs easier Partnerships are Multipliers — Technology enables, but relationships with LGUs and other government bodies scale

What's next for GabAI

At the moment, GabAI is built with PGH patients in mind, so post hackathon we would like to implement the following scaling strategies: (1) LGU partnerships—we work with local government units to contribute facility data for their cities. They have the most accurate, up-to-date facility information. (2) DOH integration—ideally, we integrate with the DOH's facility registry so we're always pulling current data. (3) Community contribution—users can flag outdated facility info, which feeds back into the database.