Dora - Customer Support Agent

A document-based AI customer support agent that answers questions using only uploaded documents.

Architecture

┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│   React UI      │────▶│  FastAPI Server │────▶│  Ollama/Mistral │
│   (Vite)        │     │  (Python)       │     │  (Local LLM)    │
└─────────────────┘     └─────────────────┘     └─────────────────┘
        │                       │
        │                       ▼
        │               ┌─────────────────┐
        │               │   RAG Module    │
        │               │  (JSON Store)   │
        │               └─────────────────┘

Tech Stack

Component	Technology
Frontend	React + Vite
Styling	Inline JSX Styles
Backend	FastAPI (Python)
LLM	Mistral via Ollama
RAG	Custom keyword search
Storage	JSON file

Project Structure

CustomerSupport/
├── server.py           # API endpoints (/upload, /chat, /documents)
├── rag.py              # Document chunking, search, context retrieval
├── system_prompt.py    # AI personality and instructions
├── requirements.txt    # Python dependencies
└── frontend/
    ├── src/
    │   ├── App.jsx         # React UI components + inline styles
    │   └── main.jsx        # React entry point + global styles
    ├── index.html
    ├── package.json
    └── vite.config.js

How RAG Works

1. Upload - Extract text from PDF/TXT/MD → Split into 800-char chunks → Store with tokenized keywords
2. Query - Tokenize question → Search all documents → Return top 3 matches per document
3. Answer - Send context + question to LLM → Stream response back to UI

Prerequisites

• Python 3.10+
• Node.js 18+
• Ollama with Mistral model

Installation

1. Install Ollama and pull Mistral:

# Install Ollama from https://ollama.ai
ollama pull mistral

2. Install Python dependencies:

cd CustomerSupport
pip3 install -r requirements.txt

3. Install frontend dependencies:

cd frontend
npm install

Running the App

1. Start Ollama (if not running):

ollama serve

2. Start the backend server:

cd CustomerSupport
python3 server.py

3. Start the frontend (new terminal):

cd CustomerSupport/frontend
npm run dev

4. Open http://localhost:3000

Usage

1. Upload one or more documents (PDF, TXT, or MD)
2. Ask questions about the uploaded content
3. Dora will answer using only information from the documents

API Endpoints

Method	Endpoint	Description
GET	/health	Check server and Ollama status
POST	/upload	Upload a document
GET	/documents	List uploaded documents
DELETE	/documents	Clear all documents
POST	/chat/stream	Send message and stream response

Features

• Document upload (PDF, TXT, MD)
• Multi-document search
• Real-time streaming responses
• Source citation in answers
• Clean, responsive UI