The central API server for the ELDORA ecosystem — handling caregiver authentication, multi-home management, IoT device heartbeats and commands, voice intent processing, FCM push notifications, scene automation, and elder analytics.
🌐 ELDORA Ecosystem · 🤖 DoraBot · 🛡️ DoraShield · 📱 Mobile App
Eldora-Backend is the central orchestration layer of the ELDORA eldercare ecosystem. It connects DoraBot hardware, the caregiver mobile app, and the AI voice pipeline into a single cohesive system — managing device registration, real-time event dispatch, command queuing, and all persistent data.
Eldora-Backend's role in the ELDORA ecosystem: "The brain — receiving signals from every device and person, deciding what matters, and making sure the right response reaches the right place at the right time."
| Runtime | Node.js (CommonJS) |
| Framework | Express 5 |
| Language | TypeScript 6 |
| ORM | Prisma 7 |
| Database | PostgreSQL |
| Auth | Firebase Auth token verification + JWT session tokens |
| Push | Firebase Admin SDK (FCM) |
| Validation | Zod v4 |
| API Docs | Swagger UI (/docs) |
| Deployment | Railway (production) |
Eldora-Backend is the central hub that all other components connect to:
ELDORA Ecosystem
├── 🛡️ DoraShield — Fall detection wearable (ESP32) → IoT events
├── 🤖 DoraBot — AI voice companion (ESP32-S3) → IoT heartbeat, voice audio, commands
├── ☁️ Eldora-Backend — API server (this repo) → orchestration layer
├── 🧠 Eldora-AI-Pipeline — Voice STT + AI + TTS service → called by backend
└── 📱 Eldora-Mobile — Caregiver app → REST API + FCM
- 🔐 Auth — Firebase ID token verification, email/password login, Google Sign-In, JWT session issuance, account deletion
- 🏠 Home — Multi-home CRUD, role-based membership (home owner / administrator / common member), invite-code invitations, emergency contacts, safety and wellness summaries, home location (latitude/longitude)
- 📡 IoT — Device self-registration on first heartbeat, telemetry intake (battery, Wi-Fi RSSI, local IP, firmware version), pending command delivery, command acknowledgment, fall event ingestion, stale device offline detection
- 🔌 Devices — Device listing, standard pairing, local token-based pairing with approval flow, pairing request management (approve/reject), room category management, Wi-Fi config command queuing, device visibility and sort management, per-device voice configuration
- 🔔 Notifications — FCM push via Firebase Admin SDK, per-user notification history, per-category notification preferences with Do Not Disturb scheduling, notification response recording, scheduled follow-up processor
- 🎬 Scenes — Full scene CRUD with five trigger types (tap-to-run, device status change, scheduled, weather change, family going home), enabled/disabled toggle, scheduled scene background executor
- 🎙️ Voice — Raw audio ingestion from DoraBot, forwarding to the AI pipeline, eleven-intent local classifier (emergency, fall, call family, comfort, etc.) with wake-phrase gating, TTS response command creation, emotion state logging (six states), voice test-speak endpoint
- 📈 Analytics — Voice interaction analytics by day and hour, emotion state breakdown per period, intent distribution, notification event analytics
| Layer | Technology | Purpose |
|---|---|---|
| Runtime | Node.js | JavaScript server runtime |
| Framework | Express 5 | HTTP routing, middleware pipeline |
| Language | TypeScript 6 | Type-safe server-side development |
| ORM | Prisma 7 + @prisma/adapter-pg |
Schema, migrations, type-safe queries |
| Database | PostgreSQL (pg) |
Primary data store |
| Authentication | Firebase Admin + jsonwebtoken |
Token verification + JWT session issuance |
| Push Notifications | Firebase Admin SDK (FCM) | Real-time push to caregiver app |
| Validation | Zod v4 | Request body and param validation |
| Password Hashing | bcryptjs | Secure credential storage |
| API Documentation | swagger-jsdoc + swagger-ui-express | Auto-generated OpenAPI spec at /docs |
| Security | Helmet | HTTP security headers |
| HTTP Client | Axios | Outbound calls to voice AI pipeline |
| Logging | Morgan | HTTP request logging |
| Dev Server | Nodemon + ts-node + tsconfig-paths | Hot-reload TypeScript development |
Eldora-Backend/
│
├── 📁 prisma/
│ ├── schema.prisma # Full database schema — all models and enums
│ ├── seed.ts # Database seed script
│ └── migrations/ # Prisma migration history
│
├── 📁 src/
│ ├── index.ts # Server bootstrap — Express app, middleware, routes, background jobs
│ │
│ ├── 📁 config/
│ │ ├── database.ts # Prisma client singleton
│ │ ├── env.ts # Environment variable loading and validation
│ │ ├── firebase.ts # Firebase Admin SDK initialization
│ │ └── swagger.ts # OpenAPI / Swagger spec configuration
│ │
│ ├── 📁 middlewares/
│ │ ├── auth.middleware.ts # JWT bearer auth (caregiver routes)
│ │ ├── device.middleware.ts # Device key auth (IoT routes)
│ │ ├── error.middleware.ts # Global Express error handler
│ │ └── logger.middleware.ts # Morgan request logger
│ │
│ ├── 📁 modules/ # Feature modules — each follows the same layered pattern
│ │ ├── 📁 auth/ # auth.routes → auth.controller → auth.service → auth.repository
│ │ ├── 📁 home/ # home.routes → home.controller → home.service → home.repository
│ │ ├── 📁 devices/ # devices + rooms + voice-config sub-controllers
│ │ ├── 📁 iot/ # iot.routes → iot.controller → iot.service → iot.repository
│ │ ├── 📁 notifications/ # notifications + follow-up processor
│ │ ├── 📁 scenes/ # scenes + scheduled executor + scene templates
│ │ ├── 📁 voice/ # audio ingestion, intent detection, emotion logging
│ │ └── 📁 analytics/ # voice and notification analytics aggregation
│ │
│ ├── 📁 shared/
│ │ ├── errors/ # AppError class and typed error helpers
│ │ ├── security/ # Local pairing token utilities
│ │ └── validations/ # Shared Zod validation schemas
│ │
│ ├── 📁 types/
│ │ ├── express.d.ts # Express Request type augmentation
│ │ └── iot.types.ts # IoT-specific type definitions
│ │
│ └── 📁 utils/
│ ├── jwt.utils.ts # JWT sign / verify helpers
│ └── response.utils.ts # Standardized API response shape
│
└── 📄 .env.example # Required environment variables template
Module layer pattern
Every module follows the same four-layer structure so the codebase is predictable:
| File | Role |
|---|---|
*.routes.ts |
Registers Express routes, applies auth middleware, maps to controller functions. Swagger JSDoc annotations live here. |
*.controller.ts |
Parses and validates the HTTP request (via Zod), calls the service, and returns the response. No business logic. |
*.service.ts |
Business logic — orchestrates repository calls, applies rules, fires side effects (FCM, command creation, etc.). |
*.repository.ts |
Direct Prisma queries. No logic beyond data access. |
The schema uses a Ms (master) / Tr (transaction) naming convention for table clarity.
MsUser — Caregivers and family members (roles: family / caregiver / admin)
MsEmergencyContact — Emergency contacts scoped to a user or home
MsNotificationPreference — Per-user, per-category notification toggles + DnD schedule
MsNotificationDeviceToken — FCM device tokens (multi-device per user)
MsHome — Elder home (name, address, lat/lng)
TrHomeMember — User ↔ Home membership (roles: home_owner / administrator / common_member)
TrHomeInvitation — Invite-code based home invitations with expiry
MsElderProfile — Elder identity linked to one or more caregivers
MsDevice — IoT device (DoraBot / DoraShield) with telemetry fields
TrDeviceCommand — Command queue (configure_wifi / activate_local_alarm / speak_on_dorabot)
TrDevicePairingRequest — Multi-caregiver pairing approval flow
MsRoomCategory — Room labels per home for device grouping
MsScene — Automation scenes with five trigger types
MsDeviceVoiceConfig — Per-device TTS voice, language, and speech rate
TrVoiceEmotionLog — Emotion state per voice interaction (neutral / calm / happy / sad / anxious / distressed)
TrNotification — Notification records (types: alarm / home / device)
TrNotificationUserState — Per-user read/unread state for multi-member homes
TrNotificationResponse — Caregiver responses to alarm notifications
In plain English:
Server starts → connects to PostgreSQL → initializes Firebase Admin
→ mounts routes: /auth /home /iot /devices /notifications /scenes /voice /analytics
→ starts three 60-second background intervals:
follow-up notification processor
scheduled scene executor
stale device offline event detector
→ handles requests:
DoraBot heartbeat → upsert device telemetry → return pending commands
DoraBot audio → forward to AI pipeline → intent classify → queue TTS command → push notification if alarm
Caregiver login → verify Firebase token → issue JWT
Caregiver app → full CRUD over homes, devices, scenes, alerts, analytics
The backend includes a local intent classifier in voice.service.ts that runs after the AI pipeline returns a transcript. This layer provides immediate intent routing without an additional LLM call.
| Intent | Trigger Keywords | Severity | Action |
|---|---|---|---|
fall_emergency |
"i fell", "fell down", "fallen" | 🔴 critical | FCM alarm + queue TTS response |
emergency_help |
"help", "emergency" | 🔴 critical | FCM alarm + queue TTS response |
call_family |
"call my family", "call my caregiver", "call someone" | 🔴 critical | FCM alarm + queue TTS response |
request_water |
"water", "drink" | 🔴 critical | FCM alarm + queue TTS response |
request_medicine |
"medicine", "medication", "pill" | 🔴 critical | FCM alarm + queue TTS response |
check_in |
"are you there", "can you hear me", "eldora" | — | Queue TTS response only |
comfort |
"lonely", "scared", "stay with me" | — | Queue TTS response only |
conversation |
(catch-all) | — | Queue TTS response only |
ignored |
No wake phrase detected | — | Optionally queue clarification |
Wake phrase gate: utterances must contain "eldora" or "el dora" unless they contain a direct emergency keyword — this prevents DoraBot from reacting to background conversation.
Emotion logging: every voice interaction logs one of six emotion states (neutral, calm, happy, sad, anxious, distressed) and a confidence score from the AI pipeline into TrVoiceEmotionLog, powering the analytics dashboard.
All routes are documented in Swagger at /docs when the server is running.
| Prefix | Auth | Description |
|---|---|---|
GET /health |
None | Liveness check — returns { status: "ok", timestamp } |
GET /docs |
None | Swagger UI — full interactive API documentation |
/auth/* |
Firebase token → JWT | Login, register, Google login, delete account |
/home/* |
JWT | Homes, members, invitations, summaries, emergency contacts |
/iot/* |
Device key | Heartbeat, fall/offline events, commands, command ACK |
/devices/* |
JWT | Device list, pair, local-pair, pairing requests, rooms, Wi-Fi queue, voice config |
/notifications/* |
JWT | Notification history, preferences, responses |
/scenes/* |
JWT | Scene CRUD, tap-to-run execution |
/voice/* |
Device key / JWT | Audio processing (device), test-speak (caregiver) |
/analytics/* |
JWT | Voice interaction and notification analytics |
Create a .env file in the project root (copy from .env.example):
NODE_ENV=development
PORT=3000
# PostgreSQL
DATABASE_URL="postgresql://user:password@localhost:5432/eldora"
# JWT session tokens
JWT_SECRET=your-long-random-secret
JWT_EXPIRES_IN=7d
# IoT device authentication
IOT_DEVICE_PROVISIONING_SECRET=your-iot-provisioning-secret
# Firebase Admin SDK — inline JSON or file path (uncomment one)
FIREBASE_SERVICE_ACCOUNT_JSON={"type":"service_account","project_id":"..."}
# FIREBASE_SERVICE_ACCOUNT_PATH=./firebase-service-account.json
# Voice AI Pipeline (Eldora-AI-Pipeline service)
VOICE_AUDIO_PROCESSOR_URL=http://localhost:8000/api/process-audio
VOICE_AUDIO_BASE_URL=http://localhost:8000
⚠️ Never commit.envorfirebase-service-account.jsonto version control. TheIOT_DEVICE_PROVISIONING_SECRETauthenticates all IoT devices — treat it as a high-value secret.
- Node.js 20+
- PostgreSQL (local or hosted)
- A Firebase project with Admin SDK credentials
# 1. Clone the repository
git clone https://github.com/Eldoraaa/eldora-backend.git
cd eldora-backend
# 2. Install dependencies
npm install
# 3. Copy and fill in environment variables
cp .env.example .env
# Edit .env with your database URL, JWT secret, Firebase credentials, and voice service URL
# 4. Run database migrations
npm run migrate:deploy
# 5. (Optional) Seed the database
npm run seed
# 6. Start the development server (hot-reload via nodemon)
npm run devThe server starts at http://localhost:3000.
Swagger UI is available at http://localhost:3000/docs.
# Compile TypeScript → dist/
npm run build
# Run compiled output
npm startnpm run build runs tsc followed by tsc-alias to resolve path aliases in the compiled output.
Database migrations
# Apply all pending migrations (development + production)
npm run migrate:deploy
# Generate Prisma client after schema changes
npm run prisma:generateThe migration history is in prisma/migrations/. Each migration file is timestamped and describes its purpose in the filename (e.g., 20260601013000_add_scenes).
ELDORA — BINUS BM Team Passage to ASEAN Hackathon 2026
| Name | Role |
|---|---|
| Stanley Nathanael Wijaya | Team Lead |
| Lutfi Alvaro Pratama | IoT Engineer |
| Andrian Pratama | Mobile Developer |
| Khalisa Amanda Sifa Ghaizani | Backend Developer |
| Devon Nicholas | AI Engineer |
Have questions, want to collaborate, or interested in ELDORA?
| Channel | Details |
|---|---|
| stanley.n.wijaya7@gmail.com | |
| @xstynwx | |
| 💬 Discord | stynw7 |
Made with 🤍 by BINUS BM Team 🔥