Compassionate support for cancer patients in Sri Lanka.

Ayu is a mental health and cancer support platform built for Sri Lankan patients. It combines an AI-powered chatbot, mood tracking, medication reminders, doctor consultations, and video recommendations into a single mobile app ,backed by two FastAPI services, a Next.js doctor dashboard, and a Next.js admin dashboard.

Tested on Pixel 9 Pro XL · Flutter 3.41.5

AI Chatbot RAG-powered support using Gemini 2.5 Flash Lite and ChromaDB. Crisis detection with safety routing.	Mood Journal Daily journaling with AI sentiment analysis (BNB, LR, XGBoost ensemble).	Medication Tracker Scheduled medication reminders with daily check-in logs.	Connect Doctor Book appointments and join Zoom consultations with oncologists.
Community Peer support community for patients and caregivers.	Companions Connect with family members to let them check on you and for emotional support.	Video Recommendations Personalised YouTube videos generated by Ollama and the YouTube API, cached per user in Redis.	Todo List Personal task management to help patients stay organised during treatment.
Donations Submit and track donation requests for treatment support.

It also includes Articles (read health articles published by admins).

Next.js web app used by oncologists to manage their patients and appointments.

• Overview stats - total patients, upcoming appointments, and completed sessions at a glance
• Calendar - monthly view of all scheduled appointments
• Appointment timeline - chronological list of upcoming consultations with patient details
• Appointment details - view intake notes, add clinical notes, upload prescriptions and documentation, and start Zoom call
• Past appointments - searchable history of completed sessions
• Patient profile - full patient record including mood stats, journal summary, and medication list

Next.js web app for platform administrators.

• Patient management - browse all registered patients and view individual profiles
• Doctor management - add, view, and manage doctor accounts
• Articles - create, edit, and publish health articles for the patient community
• Community feed - create posts and moderate the community feed
• Moderation - review and action flagged community content
• Documents - manage platform-level documents and resources
• Logs - view system and audit logs
• Edit profile - update admin account details

ayu/
├── backend/                  # Main FastAPI backend (3 Docker replicas)
├── backend-admin/            # Admin FastAPI backend (local only, no Docker)
├── frontend/
│   ├── mobile/               # Flutter patient app
│   └── web/
│       ├── doctor-dashboard/ # Next.js doctor dashboard
│       └── admin-dashboard/  # Next.js admin dashboard
├── nginx/                    # Nginx config + NJS uid-hash module
├── models/                   # ML model .pkl files
├── data/                     # Knowledge base + ChromaDB (not in git see below)
├── notebooks/                # Jupyter notebooks for model training and experimentation
│   ├── Sentiment Analysis.ipynb
│   ├── Chatbot.ipynb
│   └── Video Recommendation.ipynb
└── docker-compose.yml

data/ is not tracked in git because it contains large dataset files. You need to provide:

• data/cancer_knowledge_base.json - the RAG knowledge base used by the chatbot. This is a JSON file containing curated cancer-related information. You can build it from your own sources or request it from the team.
• data/chroma_db/ - do not create this manually. ChromaDB builds it automatically on first startup from cancer_knowledge_base.json

If you are setting up a fresh environment, the minimum you need is cancer_knowledge_base.json. Without it the chatbot will fail to initialise.

Get a Firebase service account key from Firebase Console, Project Settings, Service accounts, Generate new private key and place it in:

backend/firebase-adminsdk.json
backend-admin/serviceAccountKey.json

Download google-services.json from Firebase Console, Project Settings, Your apps, Android and place it at:

frontend/mobile/android/app/google-services.json

The main backend runs as 3 replicas behind Nginx with a shared Redis instance.

Before starting, ensure these exist:

• backend/.env
• backend/firebase-adminsdk.json
• models/ - bnb_model.pkl, lr_model.pkl, xgb_model.pkl, vectorizer.pkl, label_encoder.pkl
• data/cancer_knowledge_base.json
• FastEmbed all-MiniLM-L6-v2 model cached locally - Docker runs with HF_HUB_OFFLINE=1 so the model must be cached before the container starts or the chatbot will fail to initialise

data/chroma_db/ does not need to exist beforehand - it is created automatically on first startup.

docker compose up --build       # build and start
docker compose up --build -d    # run in background
docker compose logs -f          # stream all logs
docker compose logs -f backend1 # logs from one replica
docker compose down             # stop
docker compose down -v          # stop and clear Redis data

API available at http://localhost/api. First startup takes a few minutes while ML models load and ChromaDB builds its index.

The admin backend has no Docker setup therefore run it locally (see below).

cd backend
python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

Add to backend/.env when running without Docker:

REDIS_URL=redis://localhost:6379/0

cd backend-admin
python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
uvicorn main:app --reload --port 8001

cd frontend/web/doctor-dashboard
npm install
npm run dev   # http://localhost:3000

cd frontend/web/admin-dashboard
npm install
npm run dev -- -p 3001   # http://localhost:3001

cd frontend/mobile
flutter pub get
flutter run

APP_NAME=AYU Backend API
API_PREFIX=/api
CORS_ORIGINS=["http://localhost:3000","http://127.0.0.1:3000"]

FIREBASE_CREDENTIALS_PATH=firebase-adminsdk.json
FIREBASE_PROJECT_ID=your-project-id
FIREBASE_STORAGE_BUCKET=your-project-id.firebasestorage.app

CLOUDINARY_URL=cloudinary://api_key:api_secret@cloud_name
CLOUDINARY_CLOUD_NAME=your-cloud-name
CLOUDINARY_API_KEY=your-api-key
CLOUDINARY_API_SECRET=your-api-secret

GEMINI_API_KEY=your-gemini-api-key

OLLAMA_HOST=https://your-ollama-host
OLLAMA_CF_CLIENT_ID=your-cloudflare-access-client-id
OLLAMA_CF_CLIENT_SECRET=your-cloudflare-access-client-secret
MODEL=gemma4:e4b-it-q8_0

YOUTUBE_API_KEY=your-youtube-api-key
HF_TOKEN=your-hf-token

ZOOM_ACCOUNT_ID=your-zoom-account-id
ZOOM_CLIENT_ID=your-zoom-client-id
ZOOM_CLIENT_SECRET=your-zoom-client-secret

SENDGRID_API_KEY=your-sendgrid-api-key
SENDGRID_FROM_EMAIL=your-sender-email

ZOOM_USER_ID=your-zoom-user-id

# Optional path overrides (defaults shown)
MODEL_DIR=../models
CHROMA_DB_DIR=../data/chroma_db
KNOWLEDGE_BASE_PATH=../data/cancer_knowledge_base.json

SENDGRID_API_KEY=your-sendgrid-api-key
SENDGRID_FROM_EMAIL=your-sender-email
APP_BASE_URL=http://localhost:3001

Both the chatbot and the video recommendation service support Gemini and Ollama as interchangeable providers. Switch between them in one of two ways, no other code changes needed:

• Edit the defaults directly in backend/app/core/config.py
• Or add the variables to backend/.env to override the defaults at runtime (the .env value takes priority)

Using Gemini for everything (no Ollama required):

CHATBOT_PROVIDER=gemini
VIDEO_PROVIDER=gemini
GEMINI_API_KEY=your-gemini-api-key

Using Ollama for everything:

CHATBOT_PROVIDER=ollama
VIDEO_PROVIDER=ollama
OLLAMA_HOST=https://your-ollama-host
MODEL=gemma4:e4b-it-q8_0

The backend validates these settings at startup and will exit with a clear error if a required key is missing for the selected provider.

RividuPesara

AizaAshaf

RumethW

TashiAMFernando

MethmiDeSilva