Saathi (demo)

An AI eldercare companion — a warm, voice-first companion for the senior, and a calm status dashboard for family, both grounded in a shared "life graph."

saathi/
├── backend/    Express API — chat, routine anomaly detection, family updates
└── frontend/   Plain HTML/CSS/JS — senior companion view + family dashboard view

Run both together

1. Backend

cd backend
npm install
cp .env.example .env   # add your GEMINI_API_KEY and GROQ_API_KEY — see backend/README.md
npm start

Runs at http://localhost:3001.

2. Frontend (separate terminal)

cd frontend
npx serve .

Open the URL it prints. Use the pill switcher at the top to flip between the senior and family views — CORS is already open on the backend, so no extra config is needed for them to talk to each other locally.

If you run the backend on a different port or host, update API_BASE at the top of frontend/app.js to match.

AI provider

The backend calls Gemini first for every chat/family-update request, and automatically falls back to Groq if Gemini errors or hits a rate limit — no manual switching needed. Both have free tiers with no credit card or student ID required. See backend/README.md for setup links and details.

What's real vs. stubbed

Wired end-to-end: companion chat (with persisted, resumable, expandable conversation history), a persisted daily routine checklist with a gentle AI check-in when something's left unchecked, real voice input via the Web Speech API, a grouped family memory gallery, a two-minute box breathing exercise, a soothing-sounds player, today's routine anomaly view, family contacts, family status card, family-update narrative, and the demo day-simulation toggle — all backed by the actual API (except breathing and sounds, which are intentionally client-side only).

Stubbed (UI only, no backend yet): the senior's "I need help" button and the family app's "call now" / "send photo" buttons. Both apps' data is currently seeded from backend/lifeGraph.json and backend/todayStatus.json rather than a real device/sensor feed — see backend/README.md for what that would take to make real. The memory gallery uses placeholder photo URLs rather than a real upload flow.

See backend/README.md and frontend/README.md for details specific to each half.