cf_ai_camp_scheduler

An AI-powered summer camp schedule builder running on the Cloudflare Developer Platform.

Chat with an AI assistant (Llama 3.3 70B via Workers AI) over WebSocket to create and manage camp schedules. The AI understands natural language and uses tools to persist data in a D1 database through a Durable Object session.

---

Architecture

Component	Technology
Backend Worker	Cloudflare Workers + Hono
Real-time session	Durable Objects (WebSocket)
AI model	@cf/meta/llama-3.3-70b-instruct-fp8-fast via Workers AI
Database	Cloudflare D1 (SQLite)
Frontend	React + Vite deployed as a Cloudflare Worker

Data Model

The schema consists of six tables:

• Camp — a summer camp with dates and capacity
• Activity — a schedulable activity (sport / cultural / slow, indoor or outdoor)
• Room — a physical space with capacity and indoor/outdoor flag
• Schedule — links a set of slots to a camp
• ScheduleSlot — assigns an activity + room to a specific day (1–7) and period (morning / afternoon)
• ChatMessage — persists conversation history per camp for multi-turn AI context

---

Running Locally

Prerequisites

• Node.js >= 18
• A Cloudflare account with Workers and D1 access
• wrangler CLI authenticated (npx wrangler login)

---

Backend

cd backend
npm install

Create a .dev.vars file for local environment variables:

echo "ALLOWED_ORIGIN=http://localhost:5173" > .dev.vars

Apply the D1 migration locally:

npx wrangler d1 execute camps_database --local --file=migrations/0001_init.sql

Start the dev server:

npx wrangler dev

The backend will be available at http://localhost:8787.

Connect via WebSocket

npx wscat -c "ws://localhost:8787/ws?camp-id=1"

Replace camp-id with the ID of an existing camp. If not the AI will create one for you. You can then chat with the AI directly from the terminal:

> Create a camp called "Summer 2025" starting 2025-07-01 and ending 2025-07-14 for 30 kids
< [Tool Used: Create Camp] Camp created with id 1
< I've created the camp "Summer 2025" running from July 1st to 14th for 30 campers.

> Add a football activity, outdoor sport
< [Tool Used: Create Activity] Activity created with id 1
< Football has been added as an outdoor sport activity.

> Create a schedule for camp 1
< [Tool Used: Create Schedule] Schedule created with id 1

> Assign football on day 1 morning in the main field
< [Tool Used: Assign Slot] Slot assigned with id 1
< Football is now scheduled for Monday morning on the main field.

---

Frontend

cd frontend
npm install
npm run dev

The frontend dev server will be available at http://localhost:5173.

Make sure the backend is running on port 8787 so the frontend can connect to the WebSocket.

---

Potential Improvements

• WebSocket reconnection — automatically reconnect with backoff if the connection drops, and surface the status to the user
• Streaming AI responses — stream tokens as they arrive instead of waiting for the full reply, using Workers AI's streaming support
• Schedule conflict detection — reject assign_slot calls that double-book a room for the same day and period
• Weather-aware scheduling — integrate a weather API to prefer indoor activities on rainy days when building a schedule
• Schedule comparison — side-by-side diff view when a camp has multiple schedules
• Delete / update tools — give the AI the ability to remove or modify existing slots, activities, and rooms
• Authentication — gate the app behind Cloudflare Access or a simple auth layer so only authorized users can manage camps
• Export — download a schedule as CSV or PDF directly from the UI

---

AI Tools

The AI assistant has access to the following tools:

Tool	Description
create_camp	Creates a new summer camp
create_activity	Adds an activity (sport / cultural / slow)
create_room	Adds a room or outdoor space
create_schedule	Creates an empty schedule for a camp
assign_slot	Assigns an activity + room to a day and period
get_activities	Lists all available activities
get_rooms	Lists all available rooms
get_schedule	Retrieves the full schedule for a given schedule ID

The AI automatically decides whether to call tools based on message intent — action phrases like "create", "add", "assign" trigger tool use, while questions get direct answers.