Tabletalk

Voice AI restaurant ordering system with real-time order management.

Comment

Inspiration

Traditional restaurant ordering has friction points:

  • Long wait times for server attention
  • Miscommunication between customers and staff
  • Overwhelmed staff during peak hours
  • Impersonal digital kiosk experiences
  • Language barriers

TableTalk helps with these by providing an always-available and patient AI assistant that handles ordering while staff focus on food preparation and customer service.

Core Features

Voice AI Ordering

Customers speak naturally to Rachel, who understands context, handles modifications, and confirms orders accurately. The AI uses Google Gemini for intent parsing via an AI agent, ensuring orders are captured correctly.

Real-Time Dashboard

Restaurant staff see orders instantly as they're placed. The dashboard shows order status, queue position, wait times, and allows sending notifications to customers.

Order Lifecycle Management

Orders flow through stages: Received → Queued → Preparing → Ready → Completed. Staff update status with one click.

Smart Cart Sync

The AI automatically updates the customer's cart as they speak. Say "add two hot burgers" and watch them appear. Say "remove the shake" and it's gone.

How It Works

  1. Customer Arrives: Scans QR code or enters table code
  2. Conversation Starts: Rachel greets them and offers to help
  3. Natural Ordering: Customer speaks their order conversationally
  4. AI Processing: Gemini parses intent, updates cart in real-time
  5. Confirmation: Rachel reads back the order for verification
  6. Submission: Customer confirms, order appears on restaurant dashboard
  7. Preparation: Staff manage order through status stages

How we built it

I started with the customer experience—designing how the conversation with the AI server should feel, choosing to go with a retro pixel-art aesthetic (teal, orange, cream palette with Press Start 2P font) because it captures the heart of the diner experience which in a way is an inspiration for the experience design.

Challenges we ran into

  • Voice AI accuracy: Getting the models to reliably parse natural language into structured menu items with quantities and modifications
  • Real-time sync: Ensuring the cart updates instantly as customers speak, without lag or duplicate items [still a work in progress but better than at the start]
  • Order state management: Handling the full lifecycle from voice input → cart → confirmation → kitchen
  • Mobile responsiveness: Making voice interaction work smoothly on phones where most customers would use it [also still a work in progress]
  • Balancing AI autonomy: Rachel needed to be helpful without being annoying—confirming orders without over-confirming every small change

Accomplishments that we're proud of

  • Built a complete end-to-end voice ordering system that actually works
  • Created a restaurant dashboard that staff can realistically use during service
  • Achieved near-instant cart sync with voice input
  • Designed an interface that makes AI ordering feel more natural

What we learned

  • Voice AI requires extensive prompt engineering to handle the messiness of real speech
  • Restaurant operations have complex workflows that simple CRUD apps don't capture
  • Real-time features add significant complexity to dev flows but helps simplify user flows
  • The "last mile" of deployment (secrets, databases, cloud config) takes longer than expected
  • User testing revealed edge cases I didn't expect

What's next for Tabletalk

  • Better mobile responsiveness
  • Better real time order sync
  • Add the send customers notifications options on the restaurant's dashboard
  • Kitchen display system integration
  • Menu management interface for restaurants

Built With

Share this project:

Updates