Skip to content

misran3/asha

Repository files navigation

Asha — WhatsApp AI Assistant for Small Businesses

"Asha" means "hope" in Hindi. Meet Ash — an AI assistant that helps small businesses rise.

Built for: Nebius Build Hackathon — Sentient AI Challenge


What is Asha?

Asha is a WhatsApp-native platform that gives small home-based businesses an autonomous AI assistant. Business owners already use WhatsApp — instead of replacing it, we enhance it with Ash, an intelligent assistant that:

  • Takes orders in natural language (including Hinglish)
  • Remembers customer preferences ("your usual?")
  • Handles confirmations and order tracking
  • Sends daily summaries to business owners
  • Operates autonomously with no apps to download

No new systems to learn. Just WhatsApp.


How It Works

For Customers

  1. Text Ash on WhatsApp: "2 tiffin tomorrow, no onion"
  2. Ash asks clarifying questions and confirms details
  3. Reply "yes" to confirm — order placed

For Business Owners

  1. Set up business via WhatsApp chat with Ash
  2. Customers text Ash directly
  3. View real-time orders on the dashboard
  4. Get daily summaries from Ash
  5. Focus on the craft — Ash handles coordination

Architecture

Customer → WhatsApp → Twilio → Nebius Serverless (Multi-Agent AI)
                                        ↓
                               Supabase Database
                                        ↓
                            Dashboard (Real-time View)

Multi-Agent System

Ash isn't a single LLM call — it's a coordinated system:

  1. Conversational Agent — Natural dialogue with customers
  2. Parsing Agent — Extracts structured orders from conversation
  3. Memory Agent — Learns customer preferences over time

Tech Stack

Component Technology
Messaging Twilio WhatsApp Sandbox
Backend Nebius Serverless (Python)
LLM Nebius Token Factory (Llama 3.1 70B)
Database Supabase (PostgreSQL + Real-time)
Dashboard Next.js 14 + Tailwind CSS

Quick Start

Prerequisites

  • Twilio account with WhatsApp Sandbox enabled
  • Supabase account
  • Nebius account (Token Factory + Serverless)
  • Node.js 18+ (for dashboard)
  • Python 3.9+ (for backend)

Setup Dashboard

cd dashboard
npm install
npm run dev

Dashboard runs on http://localhost:3000

Configure Environment

Create .env.local in /dashboard:

NEXT_PUBLIC_SUPABASE_URL=your_supabase_url
NEXT_PUBLIC_SUPABASE_ANON_KEY=your_supabase_anon_key

Demo Flow

Single-Phone Demo

  1. Pre-configure: Dashboard → "Pre-configure Demo" button
  2. Subscribe: WhatsApp → "MENU"
  3. Place order: WhatsApp → "2 tiffin tomorrow, Sector 14"
  4. Watch dashboard update in real-time when customer confirms
  5. Test memory: WhatsApp → "same as last time" (Ash remembers!)

Business Onboarding Demo

  1. Reset: Dashboard → "Reset All Data"
  2. Setup: WhatsApp → "SETUP"
  3. Walk through conversational onboarding with Ash

Key Features

Implemented (Hackathon v1)

  • ✅ Natural language order processing (Hinglish support)
  • ✅ Customer memory ("your usual order")
  • ✅ Order state management (PENDING → CONFIRMED → READY)
  • ✅ 15-minute modification window
  • ✅ Real-time dashboard updates
  • ✅ Daily summary generation
  • ✅ Conversational business onboarding

Deferred (Post-Hackathon)

  • Payment verification (UPI screenshot parsing)
  • Voice message transcription
  • Multi-business support
  • Production authentication & security
  • Menu broadcasting (cost research needed)

Hackathon Alignment

Sentient AI Challenge Criteria:

Criterion How Asha Addresses It
Autonomous Presence Ash initiates proactive outreach, sends summaries unprompted
Persistent Identity Remembers customers across sessions (orders, preferences, addresses)
Real-world Presence Operates in WhatsApp — used by millions, not a demo chatbot

Problem Statement #1:

"Build agentic pipelines that reason over context, retrieve knowledge, plan next steps, and take actions."

  • ✅ Reasons over Hinglish natural language
  • ✅ Retrieves customer history, menu, pricing
  • ✅ Plans next steps (clarify? confirm? remind?)
  • ✅ Takes actions (replies, database updates, broadcasts)

Project Structure

asha/
├── dashboard/          # Next.js dashboard (read-only business view)
│   ├── app/           # App routes (dashboard, customers, orders)
│   └── lib/           # Supabase client
├── backend/           # Nebius Serverless (Python)
│   ├── webhook.py     # Twilio webhook handler
│   ├── agents/        # Multi-agent system
│   └── utils/         # LLM, database helpers
├── docs/
│   └── design/        # Full design document
└── README.md

Design Philosophy

"The dashboard is a window, not a control panel."

  • Business owner talks to Ash via WhatsApp (primary interface)
  • Dashboard shows what's happening (real-time view)
  • No buttons to click — Ash handles everything

Zero friction: If the product is "everything via WhatsApp," then business setup must also be via WhatsApp. No web forms. Just chat with Ash.


Example Conversations

Happy Path (Hinglish)

Customer: Bhai 2 tiffin kal lunch ke liye
Ash:      Namaste! 2 Regular Tiffins for tomorrow lunch, Rs 200.
          Delivery address?
Customer: Sector 14, flat 302
Ash:      Got it! Total: Rs 200. Pay to UPI: priya@paytm
          Reply OK to confirm!
Customer: Ok
Ash:      Confirmed! See you tomorrow at lunch. Thanks!

Returning Customer Memory

Customer: Hi, same as last time
Ash:      Hey Rahul! Your usual? 3 Regular Tiffins + extra sambar
          to Sector 21 flat 105? For today's lunch or dinner?
Customer: Dinner please
Ash:      Done! Rs 320. UPI: priya@paytm. Reply OK to confirm!
Customer: 👍
Ash:      Confirmed! Enjoy your dinner, Rahul!

Team

3-person hackathon team:

  • Person 1: Webhook + AI Agents
  • Person 2: Database + State Management
  • Person 3: Dashboard + Demo Controls

Built in 6 hours.


Documentation

Full design documentation: docs/design/asha-design.md

Includes:

  • Complete system architecture
  • Multi-agent design
  • 43 test cases
  • Order state machine
  • Future roadmap

License

MIT


Acknowledgments

Built for the Nebius Build Hackathon — Sentient AI Challenge.

Problem we solve: Small business owners lose orders and create friction because they can't respond fast enough or track orders reliably while managing their actual work (cooking, packing, delivering).

Solution: Ash — an AI assistant that handles the boring parts so business owners can focus on their craft.

Asha is hope. For small businesses ready to rise.

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors