DentAI - Intelligent Dental Appointment Recovery

An AI-powered system that automatically fills last-minute cancellations by intelligently contacting patients from a recall list. Built with Claude AI for decision-making and ElevenLabs for voice calls.

What It Does

When a patient cancels their dental appointment:

1. Claude AI analyzes the recall list and ranks patients by fit (treatment match, reliability, time preference)
2. The agent automatically calls the top candidate via ElevenLabs voice AI
3. If they decline or don't answer, the agent moves to the next candidate
4. When someone confirms, the slot is filled and the dashboard updates in real-time

Tech Stack

• Backend: FastAPI (Python) + Claude AI (Anthropic)
• Frontend: React + Vite + TypeScript + Tailwind CSS
• Database: Firebase Firestore (real-time sync)
• Voice: ElevenLabs Conversational AI
• SMS: Twilio (fallback)

---

Quick Start (For Judges)

Prerequisites

• Python 3.11+
• Node.js 18+
• Firebase project with Firestore enabled
• Anthropic API key (with credits)

1. Clone and Install

# Clone the repo
git clone <repo-url>
cd Dental-Thing

# Backend dependencies
python -m venv venv
./venv/bin/pip install -r requirements.txt

# Frontend dependencies
npm install

2. Configure Environment

Create a .env file:

# Required
ANTHROPIC_API_KEY=sk-ant-...

# Firebase (download serviceAccountKey.json from Firebase Console)
GOOGLE_APPLICATION_CREDENTIALS=serviceAccountKey.json

# Optional - for live voice calls
ELEVENLABS_API_KEY=...
ELEVENLABS_AGENT_ID=...
ELEVENLABS_PHONE_NUMBER_ID=...

# Optional - for SMS fallback
TWILIO_ACCOUNT_SID=...
TWILIO_AUTH_TOKEN=...
TWILIO_PHONE_NUMBER=...

3. Run Locally

# Terminal 1 - Backend
./venv/bin/uvicorn main:app --reload --port 8000

# Terminal 2 - Frontend
npm run dev

Open http://localhost:5173 to see the dashboard.

---

Demo Flow

Option A: Full Demo (with ElevenLabs)

1. Click "Trigger Cancellation" on the dashboard
2. Watch Claude rank the patients in the queue
3. The agent calls the top patient automatically
4. When they answer, the ElevenLabs voice AI handles the conversation
5. After the call, the outcome is processed and the schedule updates

Option B: Simulated Demo (without ElevenLabs)

If ElevenLabs isn't configured, you can simulate the flow:

# 1. Reset the state
curl -X POST http://localhost:8000/reset

# 2. Trigger a cancellation (Claude will rank patients)
curl -X POST http://localhost:8000/cancellation

# 3. Simulate a patient confirming
curl -X POST http://localhost:8000/call-outcome \
  -H "Content-Type: application/json" \
  -d '{"patient_name": "Podium Judge", "outcome": "confirmed"}'

Watch the dashboard update in real-time as each step happens.

---

Dashboard Components

Component	Description
Stats Bar	Shows filled slots, revenue recovered, calls made, texts sent
Cancellation Slot	The slot being filled (red = open, cyan = filled)
Agent Status	Current agent phase (idle, ranking, calling, texting)
Daily Schedule	Today's appointments with live status updates
Patient Queue	Ranked list of candidates being contacted
Activity Log	Real-time feed of all agent actions

---

API Endpoints

Endpoint	Method	Description
/	GET	Health check
/cancellation	POST	Trigger the agent to fill a cancellation
/reset	POST	Reset all state to initial demo data
/call-outcome	POST	Process call result (confirmed/declined/no_answer)
/sms-reply	POST	Process SMS reply
/webhooks/twilio-sms	POST	Twilio SMS webhook
/webhooks/elevenlabs	POST	ElevenLabs post-call webhook

---

Architecture

┌─────────────────┐     ┌──────────────────┐     ┌─────────────────┐
│   React App     │────▶│   FastAPI        │────▶│   Claude AI     │
│   (Dashboard)   │     │   (Backend)      │     │   (Brain)       │
└────────┬────────┘     └────────┬─────────┘     └─────────────────┘
         │                       │
         │                       ▼
         │              ┌──────────────────┐
         │              │   ElevenLabs     │
         │              │   (Voice Calls)  │
         │              └──────────────────┘
         │                       │
         ▼                       ▼
┌─────────────────────────────────────────────┐
│              Firebase Firestore             │
│   (Real-time sync between all components)   │
└─────────────────────────────────────────────┘

---

Mock Data

The demo uses mock patient data in agent/mock_data.py:

• 18 patients on the recall list
• Various treatments: cleaning, filling, crown, root canal, whitening
• Realistic attributes: reliability scores, time preferences, overdue cycles

The daily schedule in agent/mock_schedule.py:

• 6 appointment slots
• 1 cancellation at 2:00 PM (cleaning, $200)

---

Key Files

├── main.py                 # FastAPI server & endpoints
├── orchestrator.py         # Claude AI orchestration loop
├── agent/
│   ├── brain.py            # Scoring & decision logic
│   ├── firestore.py        # Database operations
│   ├── mock_data.py        # Patient recall list
│   ├── mock_schedule.py    # Daily schedule data
│   └── state.py            # State machine definitions
├── src/
│   ├── App.tsx             # Main React app
│   ├── lib/firestore.ts    # Frontend Firestore listeners
│   └── components/
│       └── dashboard/      # Dashboard UI components
└── tests/
    └── test_agent.py       # Backend unit tests

---

Running Tests

./venv/bin/pytest -v

All 20 tests should pass.

---

Team

Built by DentAI for the Podium AI Hackathon