Skip to content

Debanitrkl/elixir-doc-bot

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
packages
packages
 
 
.gitignore
.gitignore
 
 
README.md
README.md
 
 
package.json
package.json
 
 
vercel.json
vercel.json
 
 

Repository files navigation

Elixir DocBot

AI-powered WhatsApp medical consultation framework. Doctors run AI chatbots on WhatsApp — patients chat in any Indian language (text or voice), the AI responds in their language, and doctors see everything in English on their WordPress dashboard.

[Patient WhatsApp] <---> [Meta Cloud API]
                              |
                              v
                   [Vercel Next.js Backend]  <---> [OpenAI GPT-4o]
                     |          |         |
                     v          v         v
               [Bhashini]  [Postgres]  [WP Plugin REST Client]
                (STT/TTS/                    |
                Translate)                   v
                                      [WordPress Admin]
                                      (Doctor Dashboard)

What It Does

  1. Patient sends a voice note in Hindi on WhatsApp
  2. Bhashini ASR converts Hindi speech to Hindi text
  3. Bhashini NMT translates Hindi text to English
  4. GPT-4o generates a medical triage response (English)
  5. Bhashini NMT translates English response back to Hindi
  6. Bhashini TTS generates Hindi audio
  7. Patient receives Hindi text + audio reply on WhatsApp
  8. Doctor sees the full conversation in English on WordPress

Supports all 22 scheduled Indian languages via Bhashini (Govt of India AI translation platform).

What You Need Before Starting

Accounts & API Keys

Service What You Need Where to Get It
Vercel Account (free tier works) vercel.com
Vercel Postgres A Postgres database Vercel Dashboard → Storage → Create Database
OpenAI API key with GPT-4o access platform.openai.com/api-keys
Meta (WhatsApp) WhatsApp Business App, Phone Number ID, Permanent Access Token developers.facebook.com → Create App → WhatsApp
Bhashini User ID + API Key (free, Govt of India) bhashini.gov.in/ulca → Register → Get API credentials
WordPress A WordPress site (self-hosted or managed, PHP 8.0+) Any WordPress host

Meta WhatsApp Business Setup

  1. Go to Meta Developer Portal and create a Business App
  2. Add the WhatsApp product to your app
  3. In WhatsApp → API Setup, note your:
    • Phone Number ID (e.g., 109234567890123)
    • Permanent Access Token (generate a system user token that doesn't expire)
  4. You'll configure the webhook URL after deploying to Vercel

Bhashini Setup

  1. Register at bhashini.gov.in/ulca
  2. Go to My Account → API Keys
  3. Note your User ID and ULCA API Key

Deploy to Vercel

One-Click Deploy

Deploy with Vercel

The repo includes a vercel.json that configures the build command and output directory automatically — no manual Root Directory setting needed.

Manual Deploy

  1. Fork/clone this repo
  2. In Vercel Dashboard → New Project → Import this repo
  3. Framework Preset will be auto-detected as Next.js
  4. Add all environment variables (see below)
  5. Deploy

Environment Variables

Set these in Vercel Dashboard → Settings → Environment Variables:

Variable Description Example
POSTGRES_PRISMA_URL Vercel Postgres connection string postgresql://user:pass@host:5432/db?sslmode=require
OPENAI_API_KEY OpenAI API key sk-...
BHASHINI_USER_ID Bhashini platform user ID your-user-id
BHASHINI_API_KEY Bhashini ULCA API key your-api-key
WHATSAPP_APP_SECRET Meta App Secret (for webhook signature verification) Found in Meta App → Settings → Basic → App Secret
ENCRYPTION_SECRET 32+ character random string for PII encryption Generate with openssl rand -hex 32
DOCBOT_MASTER_SECRET Shared secret for doctor registration Generate with openssl rand -hex 24

After Deploying

  1. Push the database schema:

    cd packages/backend
npx prisma db push

    Or use npx prisma migrate deploy if you prefer migrations.

  2. Configure WhatsApp webhook:

    • Go to Meta Developer Portal → Your App → WhatsApp → Configuration
    • Webhook URL: https://your-app.vercel.app/api/webhook/whatsapp
    • Verify Token: same value as your DOCBOT_MASTER_SECRET
    • Subscribe to: messages

WordPress Plugin Setup

Install the Plugin

  1. Zip the packages/wordpress-plugin/docbot/ folder
  2. In WordPress Admin → Plugins → Add New → Upload Plugin → upload the zip
  3. Activate the plugin

Build the Admin UI (if modifying)

cd packages/wordpress-plugin/docbot/admin
npm install
npm run build

The built assets go into admin/dist/ which the plugin auto-detects.

WordPress wp-config.php (optional but recommended)

define('DOCBOT_MASTER_SECRET', 'same-value-as-vercel-env');
define('DOCBOT_ENCRYPTION_KEY', 'a-random-32-char-string-for-wp');

First-Time Setup

  1. Go to WordPress Admin → DocBot
  2. The setup wizard will appear
  3. Choose New Registration (or use an existing API key)
  4. Enter your Vercel backend URL (e.g., https://your-app.vercel.app)
  5. Enter the master secret
  6. Optionally enter WhatsApp Phone Number ID and token
  7. Click Register & Connect

Project Structure

elixir-doc-bot/
├── packages/
│   ├── backend/                          # Next.js 14 API (deployed to Vercel)
│   │   ├── prisma/schema.prisma          # Database schema
│   │   ├── src/
│   │   │   ├── app/api/
│   │   │   │   ├── webhook/whatsapp/     # WhatsApp webhook (core pipeline)
│   │   │   │   ├── doctors/register/     # Doctor registration
│   │   │   │   ├── patients/             # Patient CRUD
│   │   │   │   ├── conversations/[id]/   # Conversations + messages + doctor reply
│   │   │   │   ├── settings/             # Doctor settings
│   │   │   │   ├── stats/                # Dashboard statistics
│   │   │   │   └── auth/verify/          # API key verification
│   │   │   └── lib/
│   │   │       ├── whatsapp/client.ts    # Meta Cloud API wrapper
│   │   │       ├── bhashini/client.ts    # Bhashini ASR/NMT/TTS pipeline
│   │   │       ├── bhashini/languages.ts # 22 Indian language definitions
│   │   │       ├── openai/client.ts      # GPT-4o chat completion
│   │   │       ├── openai/system-prompts.ts # Medical AI prompt builder
│   │   │       ├── auth/middleware.ts     # API key + webhook auth + rate limiting
│   │   │       ├── db/prisma.ts          # Prisma singleton
│   │   │       └── utils/                # Encryption, audit logging
│   │   └── vercel.json
│   │
│   └── wordpress-plugin/docbot/          # WordPress Plugin
│       ├── docbot.php                    # Plugin entry point
│       ├── includes/
│       │   ├── class-docbot-admin.php    # Admin menu + React enqueue
│       │   ├── class-docbot-api-client.php # HTTP client to Vercel
│       │   ├── class-docbot-rest-api.php # WP REST proxy (14 endpoints)
│       │   ├── class-docbot-settings.php # WP options management
│       │   └── class-docbot-encryption.php # AES-256-GCM for WP
│       └── admin/
│           └── src/                      # React 18 + Vite admin app
│               ├── App.tsx               # Router + sidebar navigation
│               ├── pages/
│               │   ├── Dashboard.tsx     # Stats, recent conversations, languages
│               │   ├── Conversations.tsx # Conversation list with filters
│               │   ├── ConversationDetail.tsx # Chat UI + doctor reply
│               │   ├── Patients.tsx      # Patient list
│               │   ├── PatientDetail.tsx # Patient details + block/unblock
│               │   ├── Settings.tsx      # All doctor configurations
│               │   └── Setup.tsx         # First-time onboarding wizard
│               └── lib/api-client.ts     # WP REST API wrapper

Architecture

Multi-Doctor Support

  • Each doctor installs the WordPress plugin on their own site
  • All doctors share a single Vercel backend
  • Data is fully isolated by doctorId — doctors only see their own patients
  • Doctors are identified by WhatsApp Phone Number ID in webhooks, by API key in REST calls

Message Pipeline

Patient Voice (Hindi) ──► WhatsApp ──► Vercel Webhook
                                          │
                                    ┌─────┴─────┐
                                    │ Download   │
                                    │ Audio      │
                                    └─────┬─────┘
                                          │
                                    ┌─────┴─────┐
                                    │ Bhashini   │
                                    │ ASR (hi)   │──► Hindi Text
                                    └─────┬─────┘
                                          │
                                    ┌─────┴─────┐
                                    │ Bhashini   │
                                    │ NMT hi→en  │──► English Text
                                    └─────┬─────┘
                                          │
                                    ┌─────┴─────┐
                                    │ OpenAI     │
                                    │ GPT-4o     │──► English Response
                                    └─────┬─────┘
                                          │
                                    ┌─────┴─────┐
                                    │ Bhashini   │
                                    │ NMT en→hi  │──► Hindi Response Text
                                    └─────┬─────┘
                                          │
                                    ┌─────┴─────┐
                                    │ Bhashini   │
                                    │ TTS (hi)   │──► Hindi Audio
                                    └─────┬─────┘
                                          │
                                    Send text + audio ──► WhatsApp ──► Patient
                                          │
                                    Store both versions ──► Postgres
                                          │
                                    Doctor sees English ──► WordPress Dashboard

Security

  • WhatsApp webhooks verified via X-Hub-Signature-256 HMAC
  • Doctor API keys: SHA-256 hashed in Postgres, AES-256-GCM encrypted in WordPress
  • Patient phone numbers encrypted at rest
  • WordPress admin auth via nonce + manage_options capability
  • Rate limiting: 30 messages/min per patient
  • Medical disclaimer in every AI system prompt

Supported Languages

Assamese, Bengali, Bodo, Dogri, Gujarati, Hindi, Kannada, Kashmiri, Konkani, Maithili, Malayalam, Manipuri, Marathi, Nepali, Odia, Punjabi, Sanskrit, Santali, Sindhi, Tamil, Telugu, Urdu

Local Development

Backend

cd packages/backend
cp .env.example .env        # Fill in your API keys
npm install
npx prisma generate
npx prisma db push          # Push schema to your database
npm run dev                 # Starts on http://localhost:3000

For local WhatsApp webhook testing, use ngrok:

ngrok http 3000
# Use the ngrok URL as your webhook URL in Meta Developer Portal

WordPress Plugin Admin (React)

cd packages/wordpress-plugin/docbot/admin
npm install
npm run dev                 # Starts Vite dev server on http://localhost:5173
npm run build               # Production build into dist/

Database Schema

Table Purpose
Doctor Doctor profile, API key hash, WhatsApp credentials
DoctorSettings Specialization, system prompt, working hours, feature flags
Patient Phone, name, preferred language, blocked status
Conversation Links doctor + patient, status tracking
Message Both original language and English translation, role, type
AuditLog Doctor actions for compliance
RateLimit Per-patient and per-doctor rate limiting

Tech Stack

Component Technology
Backend Next.js 14 (App Router) on Vercel
Database Vercel Postgres (Neon) + Prisma ORM
AI OpenAI GPT-4o
WhatsApp Meta WhatsApp Business Cloud API
Translation Bhashini (Govt of India) — ASR, NMT, TTS
WP Plugin PHP 8.0+ (backend) + React 18 + Vite (admin UI)
Auth HMAC API keys, AES-256-GCM encryption, WP nonce

License

MIT

About

AI-powered WhatsApp medical consultation framework. Doctors run multilingual AI chatbots on WhatsApp, patients chat in any Indian language, doctors see everything in English on WordPress.

elixir-doc-bot.vercel.app

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages