Pulse

Pulse is a caregiving companion for families supporting a person with dementia. It helps caregivers capture daily observations, medication adherence, and spoken notes in a structured way, then turns those records into concise summaries that are easier to review before a doctor visit.

Problem Statement

Family caregivers often notice small changes in mood, sleep, appetite, memory, behaviour, and medication routines, but those observations are usually scattered across memory, messages, paper notes, or rushed conversations. By the time a medical appointment happens, important details can be forgotten, dates may be unclear, and clinicians may not get the full picture.

Pulse addresses this by giving caregivers a lightweight way to record care events as they happen and convert them into visit-ready evidence.

Solution

The app provides a mobile-first care logging workflow where caregivers can:

• Create a simple care profile for the caregiver and patient.
• Record ad-hoc voice observations and automatically transcribe them.
• Track daily wellbeing across sleep, appetite, and mood.
• Add and manage medications, including missed-dose context and voice notes.
• View previous voice notes and transcripts.
• Generate AI-assisted medical summaries for a selected date range.
• Produce summaries in English, Chinese, Malay, or Tamil.

The generated reports combine voice notes, daily wellbeing entries, and medication logs into a plain-language narrative, condensed bullet points with evidence references, and a list of items worth flagging to a doctor. The report assistant is designed to summarize caregiver observations without making clinical diagnoses.

Tech Stack

Frontend

• Next.js 14
• React 18
• TypeScript
• Tailwind CSS
• Hugeicons for UI icons
• Next.js rewrite proxy for backend API calls

Backend

• FastAPI
• Python
• Pydantic and Pydantic Settings
• Uvicorn
• Firebase Admin SDK
• Firestore for app data
• Firebase Storage support
• OpenAI Whisper for audio transcription (Multilingual)
• OpenAI GPT-5.4 for report generation

Data and AI

• Firestore collections store patients, caregivers, voice notes, daily wellbeing entries, medications, and reports.
• OpenAI Whisper can run through the API or a local Whisper model.
• GPT-5.4 generates structured JSON reports from an evidence bundle assembled by the backend.
• Mock mode is available for demo and development flows.

Repository Structure

backend/
  api/              FastAPI route modules
  models/           Internal data models
  repositories/     Firestore persistence layer
  schemas/          Request and response schemas
  services/         Business logic, transcription, reports, mock data
frontend/
  src/app/          Next.js app routes
  src/components/   Reusable UI components
  src/hooks/        Audio recording hook
  src/lib/          API client, language, and onboarding helpers
seed_data.py        Script for seeding demo data

Key Features

• Voice-first capture: caregivers can record observations without typing long notes.
• Multilingual experience: the interface and report generation support English, Chinese, Malay, and Tamil.
• Daily wellbeing check-ins: structured inputs make it easier to notice trends over time.
• Medication tracking: medication records and related voice notes become part of the report evidence.
• AI report generation: selected care data is summarized into doctor-friendly notes.
• Deployment-ready setup: frontend is prepared for Vercel and backend for Render.

Local Setup

Backend

Install dependencies:

pip install -r backend/requirements.txt

Create a root .env file with the backend settings you need:

FIREBASE_MODE=true
FIREBASE_PROJECT_ID=your-project-id
FIREBASE_STORAGE_BUCKET=your-project-id.appspot.com
FIREBASE_CREDENTIALS_JSON={}
OPENAI_API_KEY=your-openai-key
OPENAI_REPORT_MODEL=gpt-5.4
WHISPER_MODE=api
CORS_ALLOWED_ORIGINS=http://localhost:3000,http://127.0.0.1:3000

Run the API:

uvicorn backend.main:app --reload --host 0.0.0.0 --port 8000

Health check:

GET http://localhost:8000/health

Frontend

Install dependencies:

cd frontend
npm install

Create frontend/.env.local:

API_PROXY_TARGET=http://127.0.0.1:8000
NEXT_PUBLIC_MOCK_MODE=false
NEXT_PUBLIC_MOCK_DATA_JSON=

Run the app:

npm run dev

Open:

http://localhost:3000

Environment Variables

Backend

• FIREBASE_MODE: Enables Firestore-backed persistence.
• FIREBASE_PROJECT_ID: Firebase project ID.
• FIREBASE_STORAGE_BUCKET: Firebase Storage bucket name.
• FIREBASE_CREDENTIALS_JSON: Firebase service account JSON for hosted environments.
• FIREBASE_CREDENTIALS_PATH: Local path to a Firebase service account file.
• OPENAI_API_KEY: Used for AI report generation and when WHISPER_MODE=api.
• OPENAI_REPORT_MODEL: OpenAI model used for AI report generation. Defaults to gpt-5.4.
• WHISPER_MODE: api for OpenAI Whisper API or local for local Whisper.
• WHISPER_MODEL: Local Whisper model size when using local mode.
• CORS_ALLOWED_ORIGINS: Comma-separated list of allowed frontend origins.
• MOCK_MODE: Enables backend mock data responses.
• MOCK_DATA_JSON: JSON payload for mock data.

Frontend

• API_PROXY_TARGET: Backend URL used by the Next.js proxy.
• NEXT_PUBLIC_MOCK_MODE: Enables frontend mock data mode.
• NEXT_PUBLIC_MOCK_DATA_JSON: JSON payload for frontend mock data.

Deployment

Backend on Render

Use a Render Web Service with:

Build command: pip install -r backend/requirements.txt
Start command: uvicorn backend.main:app --host 0.0.0.0 --port $PORT

Recommended backend environment variables:

FIREBASE_MODE=true
FIREBASE_PROJECT_ID=...
FIREBASE_STORAGE_BUCKET=...
FIREBASE_CREDENTIALS_JSON=...
OPENAI_API_KEY=...
OPENAI_REPORT_MODEL=gpt-5.4
WHISPER_MODE=api
CORS_ALLOWED_ORIGINS=https://your-vercel-domain.vercel.app,http://localhost:3000,http://127.0.0.1:3000

Frontend on Vercel

Create a Vercel project with frontend/ as the root directory.

Set:

API_PROXY_TARGET=https://your-render-service.onrender.com

The frontend calls relative /api/* routes and relies on the Next.js rewrite proxy, so the browser does not need to call the Render backend directly.

Notes

• Generated reports are summaries of caregiver-provided observations and are not medical diagnoses.
• Report audio playback is currently disabled in the backend.
• Firestore is the active persistence layer; the SQLAlchemy database module is kept only for compatibility with older code paths.