"Asha" means "hope" in Hindi. Meet Ash — an AI assistant that helps small businesses rise.
Built for: Nebius Build Hackathon — Sentient AI Challenge
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.
- Text Ash on WhatsApp: "2 tiffin tomorrow, no onion"
- Ash asks clarifying questions and confirms details
- Reply "yes" to confirm — order placed
- Set up business via WhatsApp chat with Ash
- Customers text Ash directly
- View real-time orders on the dashboard
- Get daily summaries from Ash
- Focus on the craft — Ash handles coordination
Customer → WhatsApp → Twilio → Nebius Serverless (Multi-Agent AI)
↓
Supabase Database
↓
Dashboard (Real-time View)
Ash isn't a single LLM call — it's a coordinated system:
- Conversational Agent — Natural dialogue with customers
- Parsing Agent — Extracts structured orders from conversation
- Memory Agent — Learns customer preferences over time
| 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 |
- Twilio account with WhatsApp Sandbox enabled
- Supabase account
- Nebius account (Token Factory + Serverless)
- Node.js 18+ (for dashboard)
- Python 3.9+ (for backend)
cd dashboard
npm install
npm run devDashboard runs on http://localhost:3000
Create .env.local in /dashboard:
NEXT_PUBLIC_SUPABASE_URL=your_supabase_url
NEXT_PUBLIC_SUPABASE_ANON_KEY=your_supabase_anon_key- Pre-configure: Dashboard → "Pre-configure Demo" button
- Subscribe: WhatsApp → "MENU"
- Place order: WhatsApp → "2 tiffin tomorrow, Sector 14"
- Watch dashboard update in real-time when customer confirms
- Test memory: WhatsApp → "same as last time" (Ash remembers!)
- Reset: Dashboard → "Reset All Data"
- Setup: WhatsApp → "SETUP"
- Walk through conversational onboarding with Ash
- ✅ 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
- Payment verification (UPI screenshot parsing)
- Voice message transcription
- Multi-business support
- Production authentication & security
- Menu broadcasting (cost research needed)
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)
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
"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.
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!
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!
3-person hackathon team:
- Person 1: Webhook + AI Agents
- Person 2: Database + State Management
- Person 3: Dashboard + Demo Controls
Built in 6 hours.
Full design documentation: docs/design/asha-design.md
Includes:
- Complete system architecture
- Multi-agent design
- 43 test cases
- Order state machine
- Future roadmap
MIT
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.