Skip to content

l69d/healthbridge

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

title HealthBridge
emoji 🏥
colorFrom green
colorTo blue
sdk docker
pinned false
license mit
short_description Multilingual health literacy assistant powered by Gemma 4

HealthBridge — Multilingual Health Literacy Assistant

Gemma 4 Good Hackathon — Kaggle, May 2026 Category: Health | Model: Gemma 4 (gemma-4-31b-it)


The Problem

More than 4.5 billion people lack access to adequate healthcare. Even where clinics exist, language is a barrier — roughly 1.5 billion people speak a language that most medical resources do not cover. When someone in rural Bihar, Lagos, or rural Guatemala experiences worrying symptoms, they face:

  • No local-language health content online
  • Inability to read pill labels or prescriptions
  • No way to quickly judge if something is an emergency

The result: people either over-react (costly ER visits for colds) or under-react (dismiss a stroke because they don't recognise the signs).

The Solution — HealthBridge

HealthBridge is a free, multilingual, multimodal health literacy assistant that runs in any browser. A person can:

  1. Type their symptoms in any language — or let the app respond in their preferred language.
  2. Upload a photo — a rash, a pill bottle, a prescription, a wound, an eye condition.
  3. Receive a plain-language explanation of what may be happening, an urgency triage (Home Care / See a Doctor / Emergency), and concrete next steps — all in their chosen language.

No medical background required. No app download. Just a free API key.


Why Gemma 4?

Gemma 4 (gemma-4-31b-it) is uniquely suited for this task:

Capability Why It Matters for HealthBridge
Natively multimodal Analyses text + images in one request — pill packs, rashes, prescriptions
140+ languages Underserved communities get answers in their mother tongue
Open weights Can be deployed locally for offline clinics with no internet
Strong instruction following Structured triage output is reliable and parseable
Function calling Future extension: pull drug interaction databases, WHO guidelines

Demo Screenshots

(Run the app locally and take screenshots, or embed hosted demo GIFs here)


Features

  • Multimodal input: text description + optional image upload (JPG, PNG, WEBP, BMP)
  • Urgency triage with colour-coded badge: 🟢 Home Care · 🟠 See a Doctor · 🔴 Emergency
  • 10 supported output languages: English, Hindi, Spanish, French, Arabic, Portuguese, Bengali, Swahili, Mandarin, Tamil
  • Quick-example buttons for one-click demos
  • Graceful model fallback: tries gemma-4-31b-it first, falls back to gemini-2.0-flash automatically
  • Structured output with consistent headings for easy parsing
  • Runs entirely in the browser via Streamlit

Architecture

User input (text + optional image)
         │
         ▼
   Streamlit frontend  (app.py)
         │
         ▼
  google-genai SDK  ──►  Gemma 4 API (gemma-4-31b-it)
                         via Google AI Studio / Vertex AI
         │
         ▼
  Structured markdown response
  ┌─────────────────────────────┐
  │ ## What Might Be Happening  │
  │ ## Urgency Level            │
  │ ## Suggested Next Steps     │
  │ ## A Word of Reassurance    │
  └─────────────────────────────┘
         │
         ▼
  Urgency badge + formatted UI

Quick Start

1 — Get a Google AI API Key (free)

Visit https://aistudio.google.com/apikey and create a key. The free tier is sufficient for demos.

2 — Install dependencies

pip install -r requirements.txt

3 — Set your API key

export GOOGLE_API_KEY="your-key-here"

Or enter it directly in the sidebar when the app launches.

4 — Run

streamlit run app.py

Open http://localhost:8501 in your browser.


Usage Examples

Text-only — fever triage

"I have had a fever of 39.5°C for 3 days, a bad sore throat, no cough, and my neck glands feel swollen. I am 28 years old."

Response (Hindi selected):

## क्या हो सकता है
यह स्ट्रेप थ्रोट (बैक्टीरियल गले का संक्रमण) या मोनोन्यूक्लिओसिस हो सकता है...

## तात्कालिकता स्तर
**डॉक्टर से मिलें** — 3 दिन से अधिक बुखार और सूजी हुई ग्रंथियां...

## सुझाए गए अगले कदम
1. 24 घंटे के भीतर डॉक्टर से मिलें...

Image + text — pill identification

Upload photo of pill packet + type: "Is it safe for a 70-year-old?"

Image-only — rash analysis

Upload photo of skin rash → HealthBridge explains likely cause and urgency.


Project Structure

healthbridge/
├── app.py            # Main Streamlit application
├── requirements.txt  # Python dependencies
└── README.md         # This file

Real-World Impact

HealthBridge targets three high-impact user groups:

  1. Rural communities in South Asia, Sub-Saharan Africa, Latin America — first-generation smartphone owners who have no local health resources.
  2. Immigrant and diaspora communities — people who speak a language different from the country's healthcare system.
  3. Low-literacy users — the app can be extended with text-to-speech output (a planned next step).

Conservative estimate: Globally, 300M+ people experience a health scare annually where they cannot access language-appropriate guidance. Even a 5% reduction in delayed-care outcomes would save tens of thousands of lives yearly.


Roadmap (post-hackathon)

  • Text-to-speech output (accessibility for low-literacy users)
  • WhatsApp / Telegram bot interface (no-app-install in low-bandwidth regions)
  • Drug interaction database via function calling
  • WHO essential medicines lookup
  • Offline mode using Gemma 4 local weights (for clinics without internet)
  • Voice input via Whisper or native speech API
  • Anonymous session logging to improve triage accuracy over time

Responsible AI Notes

  • HealthBridge never diagnoses — all responses use hedged language ("may indicate", "could be").
  • Every response ends with a disclaimer recommending professional consultation.
  • No user data is stored by the app. Image/text is sent to the Google AI API and immediately discarded.
  • The system prompt explicitly instructs the model to be culturally sensitive and avoid assumptions about healthcare access or resources.

License

Apache 2.0 — same as Gemma 4.


Author

Built by Karthik Gowda for the Gemma 4 Good Hackathon (May 2026).

"The best medicine is knowledge in the right language at the right time."


HealthBridge is an educational tool, not a substitute for professional medical advice. Always consult a qualified healthcare provider for medical decisions.

About

Multilingual health literacy assistant powered by Gemma 4 — Kaggle Gemma 4 Good Hackathon submission

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors