// Live Demo

AI AGENT
SYSTEM DEMO

A production multi-tenant AI agent platform with RAG, intelligent handoffs, and booking automation. Try the live demo below — it's trained as "Rapid Flow Plumbing" to showcase real-world capabilities.

Built by Alex Turner · Self-hosted on AWS

// Interactive Demo

Talk to the Agent

This is a fully functional AI assistant trained as a fictional plumbing business. Ask about services, pricing, or try booking an appointment.

What to Try

  • Ask about pricing for drain cleaning or water heater repairs
  • Check if they offer 24/7 emergency service
  • Try booking an appointment (the bot will collect your details)
  • Ask a question outside its knowledge to see the human handoff

How It Works

The agent uses RAG (Retrieval-Augmented Generation) with Gemini embeddings stored in Supabase's pgvector. It searches a knowledge base of 16 entries about "Rapid Flow Plumbing" services, then generates responses using Claude/GPT/Gemini via dynamic LLM routing.

Real System Features

  • Multi-tenant architecture (each client gets isolated data)
  • Rate limiting (20 messages per hour per customer)
  • Audit logging (every message, booking, handoff tracked)
  • Conversation memory (stored as JSONB in PostgreSQL)

Rapid Flow Plumbing Assistant

Status: ● ONLINE

Hi! I'm the AI assistant for Rapid Flow Plumbing. Ask me about our services, pricing, or schedule an appointment.

// Architecture

Built With

FastAPI
Python Backend
Supabase
PostgreSQL + pgvector
LangGraph
Agentic Workflows
n8n
Workflow Orchestration
Docker
Containerization
AWS Lightsail
Hosting

Multi-LLM Support: The system dynamically routes between Claude (Anthropic), GPT-4 (OpenAI), Gemini (Google), and Llama 3.3 (via Groq) based on client configuration. RAG embeddings use Gemini's 768-dimensional model for semantic search. Full source available on GitHub.