📌 GenAI Credit Card Assistant — Voice + RAG + EMI Planner + Actions

⚠️ Important Notice: The folder structure, setup instructions, and code paths detailed in this README correspond to the master branch of this repository.

A production-grade GenAI chatbot built for the FPL Technologies (OneCard) Product Builder Intern Assignment. This assistant understands voice + text queries, performs credit-card actions, explains EMI options, retrieves info using RAG, and offers a premium glass-morphism UI.

---

🚀 Key Features

🔊 1. Real-time Voice Assistant (Browser Native)

• Uses Google Web Speech API for free, low-latency speech-to-text.
• Seamless push-to-talk / stop-talk control.
• Bot responds in text and optional text-to-speech.

🤖 2. Smart Agent with Intent Classification

Understands:

• Billing queries
• Card block / unblock
• EMI conversion
• Repayments
• Delivery
• Transaction disputes
• General onboarding queries

Powered by Groq LLaMA-3.1-8B (instant) using:

• Intent classifier
• Parameter extractor
• Tool execution layer
• Rewritten queries for better understanding

📚 3. RAG (Retrieval Augmented Generation)

• Embedding model: Snowflake Arctic embed
• Vector DB: In-memory cosine search
• Handles all informational queries:
  • Bill generation rules
  • Limits
  • Fees
  • Markup
  • Eligibility
  • Refund timelines
  • Delivery stages
  • OTP, KYC, activation
• Clean inline citations using:

  [
    { "chunk": "...", "score": 0.87 },
    ...
  ]

💳 4. EMI Planner Engine (Custom-built)

• Computes EMI for 3, 6, 9, 12 months
• Includes:
  • Interest
  • GST on interest
  • Total payable
  • Recommended plan (lowest total cost)
• Formats clean readable tables

⚙ 5. Tool Execution Layer

Implements real actions (mocked):

Action	Tool
Block card	tool_block_card
Unblock card	tool_unblock_card
Get bill summary	tool_get_bill_summary
Convert to EMI	tool_convert_to_emi
Get current balance	tool_get_current_balance

🎨 6. Premium OneCard-Style UI

Built using Next.js + Tailwind + Glassmorphism

• Components:
  • Chat bubbles
  • Suggested prompts
  • Voice mic button
  • EMI info cards
• Intent + metadata debugging (optional)
• Fully responsive

---

🏗 Architecture Overview (Refined)

                               ┌──────────────────────────┐
                               │    Frontend (Next.js)    │
User ↔ Browser ↔ Voice/Input → │ - Voice input            │ ↔ REST / WebSocket
                               │ - Chat UI                │
                               │ - EMI Cards              │
                               └───────────┬──────────────┘
                                           │
                                           ▼
                                 ┌──────────────────────┐
                                 │    FastAPI Backend   │
                                 │  /chat               │
                                 │  /ws (optional)      │
                                 └───────────┬──────────┘
                                             │
                       ┌─────────────────────┴──────────────────┐
                       ▼                                        ▼
             ┌──────────────────────┐                 ┌────────────────────────────┐
             │     Agent Brain      │                 │       RAG Engine           │
             │ - Intent classify    │                 │ - Vector DB search         │
             │ - Query rewrite      │                 │ - Snowflake embeddings     │
             │ - Parameter extract  │                 │ - Top-k retrieval          │
             │ - Action routing     │                 └────────────────────────────┘
             └──────────────────────┘
                                             ▼
                                 ┌──────────────────────────┐
                                 │      Tools Layer         │
                                 │ - Block card             │
                                 │ - Bill summary           │
                                 │ - EMI computation        │
                                 │ - Balance lookup         │
                                 └──────────────────────────┘

Folder Structure

⚠️ Note: This folder structure is based on the master branch. ``` ...

genai-credit-card-assistant/
│
├── backend/
│   ├── main.py               # FastAPI server
│   ├── agents/
│   │   ├── agent.py          # Main agent logic
│   │   ├── tools.py          # Action tools
│   │   ├── emi_engine.py     # EMI calculator
│   │   └── query_rewriter.py
│   ├── rag.py                # Vector search + embeddings
│   ├── data/
│   │   └── knowledge_base.json
│   └── .env.example
│
├── voice-frontend/
│   ├── app/
│   │   └── page.tsx          # UI + Voice handling
│   └── components/
│       └── ChatBubble.tsx
│
└── README.md

🛠 Setup Instructions

1️⃣ Backend Setup

cd backend python -m venv venv

On Windows

venv\Scripts\activate

On macOS/Linux

source venv/bin/activate

Requirements

pip install -r requirements.txt

Create .env:

GROQ_API_KEY=your_key
MODEL_NAME=llama-3.1-8b-instant

Run server:

Bash: uvicorn backend.main:app --reload

2️⃣ Frontend Setup

Bash: cd voice-frontend npm install npm run dev Visit: 👉 http://localhost:3000

🧪 Example Queries

✔ Informational

“What is foreign currency markup?”

“Refund kab aata hai?”

“Delivery delay kya hota hai?”

“Minimum due kya hai?”

✔ Actions

“Block my card”

“Convert 8000 to EMI”

“What is my bill summary?”

“What’s my current balance?”

✔ Rewritten Queries (Agent auto-corrects)

“8000 ko EMI me daal do”

“kitna outstanding hai”

“due kitna hoga”