AgentAI

An AI-powered automation agent built with LangChain and Llama 2, featuring a FastAPI backend and Streamlit frontend.

Overview

AgentAI is a versatile automation platform that combines large language models with task automation capabilities. The system processes natural language requests, generates suggestions, and executes automated workflows while maintaining a conversation history.

Features

• Natural Language Processing: Powered by Meta's Llama-2-13b model
• Task Automation: Workflow execution using Prefect
• Data Validation: Built-in data validation using Great Expectations
• Conversation Logging: Automatic logging of all interactions
• Web Interface: User-friendly Streamlit frontend
• REST API: FastAPI-based backend services

Architecture

• Frontend: Streamlit application (frontend/app.py)
• Backend:
  • FastAPI server (backend/main.py)
  • LLM integration (backend/agent.py)
  • Task automation (backend/tasks.py)
  • Data monitoring (backend/monitoring.py)

Setup

Prerequisites

• Docker
• NVIDIA GPU with CUDA support
• Docker Compose

Installation

1. Clone the repository:

git clone <repository-url>
cd AgentAI

2. Configure environment:

   • Update backend/config.json with your Hugging Face token
   • Adjust model parameters if needed

3. Build and run using Docker Compose:

docker-compose up --build

Usage

Web Interface

Access the Streamlit interface at http://localhost:8501:

• Run tasks using natural language
• Validate data files
• View conversation history

API Endpoints

• GET /run-task: Execute automation tasks
• GET /validate-data: Validate data files
• GET /conversations: Retrieve conversation history

Development

Development Environment

The project includes a devcontainer configuration for VS Code:

• CUDA-enabled development environment
• Pre-configured Python extensions
• Automatic dependency installation

Project Structure

AgentAI/
├── backend/
│   ├── agent.py      # LLM integration
│   ├── config.json   # Configuration
│   ├── main.py      # API endpoints
│   ├── monitoring.py # Data validation
│   └── tasks.py     # Automation workflows
├── frontend/
│   └── app.py       # Streamlit interface
├── data/
│   └── conversations.jsonl # Conversation logs
└── docker/
    └── ...          # Docker configurations

Run Backend:

• In VSCode terminal: uvicorn backend.main:app --host 0.0.0.0 --port 8000
• Test: curl http://localhost:8000/run-task?task=process%20my%20files

Run Frontend:

• New terminal: streamlit run frontend/app.py --server.port 8501
• Access via browser (forward port 8501 via SSH if needed).

License

MIT License

Copyright (c) 2025 Shiwen An