Smart-Cart Upsell & Negotiation Agent

Razorpay AI Buildathon — Track 1 & AI Risk Track
A deterministic, auditable FinTech AI agent with zero LLM hallucinations in pricing, complete with a React/Vite Storefront and Razorpay Webhook integration.

┌─────────────────┐    chat      ┌──────────────────────────────────────┐
│  React / Vite   │ ───────────► │           FastAPI Backend            │
│  (Storefront)   │ ◄─────────── │                                      │
└─────────────────┘  pay link    │  ┌──────────┐   ┌─────────────────┐ │
                                 │  │  Groq    │   │  Financial Core │ │
                                 │  │ LLM Agent│──►│  (guardrails)   │ │
                                 │  │ (router) │   └────────┬────────┘ │
                                 │  └──────────┘            │          │
                                 │                          ▼          │
                                 │  ┌──────────┐   ┌─────────────────┐ │
                                 │  │ Supabase │◄──│   Razorpay API  │ │
                                 │  │ (DB+Audit│──►│  (Payment link) │ │
                                 │  └──────────┘   └─────────────────┘ │
                                 └──────────────────────────────────────┘

Key Features:

• Zero LLM Math: The LLM only emits JSON tool-calls. It never calculates prices or generates links.
• Razorpay Webhooks: Automatically catches payment.captured events to close the payment loop.
• Inventory Control: Deducts stock on successful payment and prevents the AI from selling out-of-stock items (Graceful Failure).
• Competitor Scraper Recovery: If the external DuckDuckGo price scraper fails, the backend catches the timeout and instructs the LLM to gracefully pivot the conversation without crashing.
• Admin Dashboard: Brutalist admin portal to monitor live revenue, sessions, and insert new inventory.

1. Go to your Supabase project's SQL Editor.
2. Run the sql/create_tables.sql script to create the schema.
3. Run the sql/seed_data.sql script to populate the catalog with test products.

Open a terminal and navigate to the project root:

# Install dependencies
pip install -r requirements.txt

# Configure environment
cp .env.example .env
# Fill in your keys (Supabase, Groq, Razorpay) in the .env file

# Run the backend
uvicorn backend.main:app --port 8000

Open a second terminal and navigate to the web folder:

cd web
npm install
npm run dev

• Storefront: Go to http://localhost:5173 to interact with the AI, negotiate a price, and buy a product.
• Admin Dashboard: Go to http://localhost:5173/admin (Passcode: smartcart2026) to view live revenue, audit logs, and add new inventory.

1. Chat with the agent on the Storefront and ask for a 4K Monitor.
2. Negotiate a discount.
3. Click the generated Razorpay link and pay (using test cards).
4. Go to the Admin Dashboard to watch your revenue increase.
5. Go back to the chat and ask to buy the same monitor again. The AI will gracefully inform you it is out of stock!