TechNova CX — An Enterprise-Grade, Air-Gapped AI Customer Experience Platform with Multi-Agent Guardrails & Human-in-the-Loop Escalation.

💡 The Problem vs. Our Solution

Traditional customer support chatbots rely on generic, single-prompt architectures that frequently hallucinate company policies, trap users in frustrating automated loops, and offer zero transparency for human support teams.

Our Solution: TechNova CX replaces generic chatbots with a self-correcting, 3-agent AI pipeline. It verifies every factual claim before communicating with the customer. The Orchestrator routes intent, the Support Agent calls live database tools, and the Critic Agent acts as a strict 0.6 confidence guardrail—seamlessly escalating complex issues to human agents with pre-drafted AI solutions.

🧠 Core Architecture Diagram

graph TD
    A[Vercel Frontend] --> B{Supabase Edge Function / API}
    B --> C[Orchestrator Agent - Gemini Flash-Lite]
    C --> D[Support Agent - Gemini Flash]
    D <-->|Execute Tools| E[(Supabase pgvector RAG)]
    D --> F{Critic Agent - Gemini Flash}
    F -- Confidence >= 0.6 --> G[Output to Customer]
    F -- Confidence < 0.6 --> H[Human Ticket Escalation]
    H --> I[Agent Assist Admin Portal]

Loading

🛠️ Local Setup Instructions

1. Clone the repository

git clone https://github.com/your-username/technova-cx.git
cd technova-cx

2. Configure Environment Variables Copy .env.example to .env.local and add your secure keys:

GEMINI_API_KEY="your_api_key_here"
NEXT_PUBLIC_SUPABASE_URL="https://your-project.supabase.co"
NEXT_PUBLIC_SUPABASE_ANON_KEY="your_anon_key_here"
SUPABASE_SERVICE_ROLE_KEY="your_service_role_key_here"

3. Install Dependencies & Initialize Database Install node modules and use the Supabase CLI to push the schemas and seed data:

npm install
supabase start
supabase db reset
# Or manually run supabase/migration.sql and supabase/seed.sql in your SQL editor

4. Start the Development Server

npm run dev

Navigate to http://localhost:3000 to experience the TechNova CX storefront.

🎯 Hackathon Demo Guide

Judges and evaluators can test our complete multi-agent pipeline using the 4-step script below:

1. Test Tool Execution (Order Tracking): On the /support page, click the quick-reply pill 📦 Where is order ORD-7734?. The Support Agent will securely fetch live order status from the Postgres table.
2. Test Grounded RAG & Citations (Warranty): Click 🛡️ Is NovaBook Pro X15 under warranty?. Watch as the Support Agent retrieves context and generates a citation chip. Click the chip to deep-link directly to the exact verified text on the /policies trust center page.
3. Test the Critic Agent Trace: After any AI response, open the ⚡ View AI Trace drawer in the chat bubble. You can actively inspect the Orchestrator's intent classification, the Support Agent's tool calls, and the Critic Agent's verification score in real-time.
4. Test Human-in-the-Loop Escalation: Click ⚠️ I need to speak to a human manager.. The Critic or Orchestrator will instantly flag the hard trigger and escalate. Navigate to /admin/tickets to see the newly escalated ticket waiting in the Agent Assist Portal with full trace context.