Kinetic Canteen - AI Dim Sum Waiter

An agentic RAG-powered dim sum waiter for Kinetic Canteen, Guelph, Ontario. Guests chat with the Dim Sum Enthusiast, an AI waiter that recommends dishes, manages a live cart, and wraps up orders with a local craft beer pairing.

Features

• Conversational ordering - natural language chat to browse the menu, add/remove items, and manage the cart
• Agentic RAG - semantic menu search via Pinecone; conversation memory via FAISS
• Live cart - Supabase Realtime syncs cart state between the AI and the cart drawer instantly
• Order completion - structured order summary, Guelph brewery beer pairing, and automatic cart clear
• Session memory - conversation summarized every 3 turns to keep context concise
• Responsive UI - Next.js 14 App Router with Tailwind CSS, brand theme Guelph Gold & Steam

---

Tech Stack

Layer	Technology
Frontend	Next.js 14 (App Router), Tailwind CSS, Supabase JS
Backend	FastAPI, LangGraph, LangServe
LLM	OpenAI gpt-4o-mini + text-embedding-3-small
Vector DB	Pinecone (serverless, menu search)
Memory	FAISS (in-process, conversation summaries)
Database	Supabase (cart_items table + Realtime)

---

Project Structure

AI Waiter/
|-- backend/
|   |-- app/
|   |   |-- main.py              # FastAPI app + LangServe mount
|   |   |-- config.py            # Settings (pydantic-settings)
|   |   |-- graph/
|   |   |   |-- state.py         # AgentState TypedDict
|   |   |   |-- nodes.py         # agent, summarizer, signature_move nodes
|   |   |   +-- workflow.py      # StateGraph assembly + routing
|   |   |-- tools/
|   |   |   |-- menu_search.py   # Pinecone semantic search tool
|   |   |   |-- memory_search.py # FAISS conversation memory tools
|   |   |   +-- cart.py          # Supabase cart tools (add/remove/get/clear)
|   |   |-- prompts/
|   |   |   +-- persona.py       # Dim Sum Enthusiast system prompt
|   |   +-- data/
|   |       +-- seed_pinecone.py # Seeds 52+ menu items into Pinecone
|   |-- requirements.txt
|   +-- .env.example
+-- frontend/
    |-- src/
    |   |-- app/
    |   |   |-- page.tsx             # Redirects to /table/3
    |   |   +-- table/[id]/page.tsx  # Table page
    |   |-- components/
    |   |   |-- ChatInterface.tsx    # Main layout + order state
    |   |   |-- CartDrawer.tsx       # Slide-in cart panel
    |   |   |-- MessageBubble.tsx    # Chat message + menu item cards
    |   |   |-- MenuItemCard.tsx     # Add-to-cart card from AI response
    |   |   |-- SuggestionChips.tsx  # Quick-reply chips
    |   |   |-- ChatInput.tsx        # Message input
    |   |   +-- TableHeader.tsx      # Header with cart badge
    |   |-- hooks/
    |   |   |-- useChat.ts           # Chat state, streaming, suggestions
    |   |   +-- useCart.ts           # Cart state + Supabase Realtime
    |   +-- lib/
    |       |-- api.ts               # Backend streaming API client
    |       |-- supabase.ts          # Supabase client
    |       +-- types.ts             # Shared TypeScript types
    |-- .env.local.example
    +-- package.json

---

LangGraph Workflow

START
  -> agent_node (GPT-4o-mini + system prompt)
       |-- tool_calls?     -> tool_node -> agent_node  (loop)
       |-- order_complete? -> signature_move_node -> END
       |-- every 3 turns?  -> summarizer_node -> END
       +-- otherwise       -> END

Nodes:

• agent_node - main conversational agent with all tools bound; detects order-complete phrases
• tool_node - executes search_menu, search_conversation_memory, add_to_cart, remove_from_cart, get_cart, clear_cart_tool
• summarizer_node - condenses conversation every 3 turns; stores summary in FAISS and trims message history
• signature_move_node - fetches the real cart from Supabase, generates order confirmation + Guelph beer pairing, clears the cart

---

Setup

Prerequisites

• Python 3.11+
• Node.js 18+
• OpenAI API key
• Pinecone account (serverless index named dim-sum-menu, dimension 1536)
• Supabase project

1. Supabase Table

Run in the Supabase SQL editor:

create table cart_items (
  id uuid primary key,
  table_id text not null,
  item_name text not null,
  quantity integer not null default 1,
  unit_price numeric(6,2) not null,
  created_at timestamptz not null default now()
);

create index idx_cart_table on cart_items(table_id);

Enable Realtime on cart_items in the Supabase dashboard under Database > Replication.

2. Backend

cd backend
python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate
pip install -r requirements.txt
cp .env.example .env
python -m app.data.seed_pinecone
uvicorn app.main:app --reload --port 8000

backend/.env

OPENAI_API_KEY=sk-...
PINECONE_API_KEY=pcsk_...
PINECONE_INDEX=dim-sum-menu
SUPABASE_URL=https://<project>.supabase.co
SUPABASE_KEY=<anon or service role key>

3. Frontend

cd frontend
npm install
cp .env.local.example .env.local
npm run dev

frontend/.env.local

NEXT_PUBLIC_SUPABASE_URL=https://<project>.supabase.co
NEXT_PUBLIC_SUPABASE_ANON_KEY=<anon key>
NEXT_PUBLIC_API_URL=http://localhost:8000

Open http://localhost:3000 — the app redirects to /table/3 by default.

---

API Reference

Method	Path	Description
POST	/chat/stream	Stream the waiter agent response (SSE)
POST	/chat/invoke	Invoke the waiter agent (blocking)
GET	/api/table/{id}/cart	Get cart items for a table
DELETE	/api/table/{id}/cart/{item_id}	Remove a cart item by ID
POST	/api/table/{id}/clear	Clear cart and conversation memory
GET	/health	Health check

---

Menu

52+ items across 10 categories: Har Kau & Siu Mai, Bao & Buns, Rice & Noodles, Cheung Fun, Turnip Cake & Taro, Soup Dumplings, Spring Rolls & Crispy, Desserts, Beverages, and Seasonal Specials. Each item is embedded and stored in Pinecone for semantic search.

---

Brand

Guelph Gold & Steam

Token	Value
Primary	#E83946 (red)
Secondary	#FF8800 (orange)
Tertiary	#FFD700 (gold)
Neutral	#121212 (near-black)