Skip to content

Latest commit

 

History

History
201 lines (155 loc) · 4.38 KB

README.md

File metadata and controls

201 lines (155 loc) · 4.38 KB
Raw

AI Sales Agent Backend

A sophisticated Node.js application that implements an AI-powered sales agent using Together AI's LLM capabilities. The platform handles product inquiries, demo scheduling, and customer interactions with a focus on AWS product information.

🌟 Features

Core Capabilities

  • 🤖 AI-powered conversational agent using Together AI's LLM (Meta-Llama-3.1-70B-Instruct-Turbo)
  • 📝 RAG (Retrieval Augmented Generation) for accurate product information retrieval
  • 📅 Automated demo scheduling with email confirmation
  • 💾 MongoDB integration for persistent data storage
  • 🔄 Real-time communication via WebSocket
  • 📊 Sophisticated conversation state management
  • 🔍 Vector search using Pinecone for semantic similarity
  • 📝 Comprehensive logging system

Technical Features

  • TypeScript-based architecture
  • Express.js server with WebSocket support
  • Modular service architecture
  • Comprehensive error handling
  • Structured logging with Winston
  • Zod schema validation
  • Environment-based configuration

🏗 Architecture

Core Components

  1. WebSocket Manager: Handles real-time communication
  2. Conversation Manager: Manages conversation flow and state
  3. Node Service: Implements conversation flow logic
  4. AI Service: Integrates with Together AI
  5. Embedding Service: Manages vector embeddings with Pinecone
  6. Product Service: Handles product-related queries

Data Models

  • Customer
  • Message
  • Event
  • Session

🚀 Prerequisites

  • Node.js (v14 or higher)
  • MongoDB (v4.4 or higher)
  • npm or yarn package manager
  • Pinecone account
  • Together AI API key

⚙️ Environment Variables

Create a .env file with:

API Keys

TOGETHER_API_KEY=your_together_ai_key PINECONE_API_KEY=your_pinecone_key PINECONE_ENVIRONMENT=your_pinecone_environment

Database

MONGODB_URI=your_mongodb_connection_string

Server

PORT=3000 LOG_LEVEL=debug

📦 Installation

  1. Clone the repository:
git clone https://github.com/gooduru-vineeth/sales-agent-backend.git
cd sales-agent-backend
  1. Install dependencies:
npm install
  1. Build the project:
npm run build
  1. Load AWS product data:
npm run load-aws-data
  1. Start the server:
npm start

🛠 Development

Available Scripts

  • npm start: Run the production server
  • npm run dev: Run development server with hot reload
  • npm run build: Build the TypeScript project
  • npm run lint: Run ESLint
  • npm run format: Format code with Prettier
  • npm run load-aws-data: Load AWS product data into Pinecone

Project Structure

src/
├── config/         # Configuration files
│   ├── database.ts
│   └── index.ts
│
├── models/         # MongoDB models
│   ├── Customer.ts
│   ├── Event.ts
│   └── Message.ts
│
├── providers/      # External service providers
│   ├── together/
│   └── pinecone/
│
├── repositories/   # Data access layer
│   ├── Customer.ts
│   ├── Event.ts
│   └── Message.ts
│
├── scripts/        # Utility scripts
│   └── loadAwsProductData.ts
│
├── services/       # Business logic
│   ├── AIService.ts
│   ├── ConversationManager.ts
│   ├── EmbeddingService.ts
│   └── NodeService.ts
│
├── types/          # TypeScript types
│   ├── ai-provider.ts
│   ├── analysis.ts
│   ├── customer.ts
│   ├── events.ts
│   └── message.ts
│
├── utils/          # Utility functions
│   └── logger.ts
│
└── websocket/      # WebSocket handling
    └── WebSocketManager.ts

🔄 Conversation Flow

The system implements a node-based conversation flow:

  1. Welcome Node
  2. Collect User Information
  3. Product Information
  4. Q&A Node
  5. Demo Scheduling
  6. End Conversation

📚 AWS Product Coverage

Currently supports information about:

  • Amazon EC2
  • Amazon S3
  • Amazon RDS
  • Amazon DynamoDB
  • AWS Lambda

🔒 Security

  • Secure WebSocket connections
  • Environment-based configuration
  • Input validation using Zod
  • MongoDB security best practices
  • API key management

📝 Logging

Comprehensive logging system using Winston with:

  • Console output
  • File-based logging
  • Error tracking
  • Request/Response logging
  • Performance monitoring

📄 License

MIT License