Riverstone Agent

A voice-enabled AI real estate sales agent for Harbourline Developments — built to qualify leads, answer property questions, and book appointments for their Melbourne apartment projects.

Live Demo

👉 Streamlit Frontend

---

Overview

This project simulates a conversational AI sales agent for four apartment developments in Melbourne:

• Riverstone Place (Abbotsford)
• Harbourview Towers (Richmond)
• Yarra Edge (Footscray)
• Collingwood Quarter (Collingwood)

The agent can:

• Qualify buyers based on budget, bedrooms, timeframe, and finance status
• Recommend the best matching project
• Handle common questions and objections
• Book mock appointments
• Log leads with compliance tracking

---

Tech Stack

• Backend: FastAPI (voice_agent.py)
• Frontend: Streamlit (app.py)
• LLM: Google Gemini (primary) with Mistral fallback
• Text-to-Speech: gTTS (frontend) + ElevenLabs (optional)
• Database: SQLite (lead logging)
• Deployment: Render (backend) + Streamlit Cloud (frontend)

---

Features

• ✅ Intelligent project recommendation based on buyer needs
• ✅ Natural conversation flow with chat history
• ✅ Appointment booking logic with available time slots
• ✅ Compliance handling (stop/unsubscribe requests)
• ✅ Lead logging with qualification data
• ✅ Voice output using gTTS (reliable on cloud)
• ✅ Rate limiting and basic security

---

Project Structure

mavrk7-riverstone-agent/
├── app.py                    # Streamlit frontend
├── voice_agent.py            # FastAPI backend
├── requirements.txt
├── Dockerfile
├── Procfile
├── .env.example
├── tests/
│   ├── llm_tester.py
│   └── test_mistral.py
└── .github/workflows/        # CI/CD pipelines

---

Local Setup

1. Clone the repository:

   git clone https://github.com/MAvRK7/riverstone-agent.git
   cd riverstone-agent
   

2. Create and activate a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows, use `venv\Scripts\activate`
   

3. Install dependencies:

   pip install -r requirements.txt

4. Create a .env file with the following content:

   GEMINI_API_KEY=your_gemini_api_key
   MISTRAL_API_KEY=your_mistral_key
   BACKEND_URL=https://your-deployed-backend-url/call
   BACKEND_API_KEY=optional_backend_api_key
   

5. Run the application:

   python voice_agent.py
   

6. Run the Streamlit frontend:

   python -m streamlit run "app.py"
   

---

Limitations (Free Demo)

• No real phone calling (Twilio/Retell not integrated)
• Backend on Render free tier → sleeps after 15 minutes of inactivity
• Voice input not implemented (planned)
• TTS uses gTTS on frontend for reliability

---

Cost & Next Steps

Cost to run ~100 calls: ~$30–50/month (Twilio for calls + persistent hosting + TTS API usage). If extended, improvements could include:

• Persistent hosting (Render paid / AWS / GCP).
• Fully integrated TTS in the cloud.
• Phone-call support via Twilio or Retell.
• Richer UI/UX with conversation history.
• Integration with Calendly or Google Calendar for real bookings.

---

Model Output

---

License

This project is for demonstration purposes. See LICENSE for details.

Made with ❤️ for showcasing AI in real estate sales

---

✅ This demo was built entirely free of cost to show end-to-end flow, despite limitations.