AI-agents-with-AutoGen

AutoGen is a framework that enables development of LLM applications using multi-agents. AutoGen framework can be used with any model via API or locally within your own environment.

Lessons

• Lesson 1: Multi-Agent Conversation and Stand-up Comedy
• Lesson 2: Sequential Chats and Customer Onboarding
• Lesson 3: Reflection and Blogpost Writing
• Lesson 4: Tool Use and Conversational Chess
• Lesson 5: Coding and Financial Analysis
• Lesson 6: Planning and Stock Report Generation

Table of contents

• AI-agents-with-AutoGen
  • Lessons
  • Table of contents
  • Directory structure
  • Getting started
  • Example of code
  • References

Directory structure

AI-agents-with-AutoGen/
├── 1 Multi-agent conversation and stand-up comedy
│   ├── L1_Multi-Agent_Conversation_and_Stand-up_Comedy.ipynb
│   ├── README.md
│   ├── requirements.txt
│   └── utils.py
├── 2 Sequential chats and Customer Onboarding
│   ├── L2_Sequential_Chats_and_Customer_Onboarding.ipynb
│   └── README.md
├── 3 Reflection and Blogpost Writing
│   ├── L3_Reflection_and_Blogpost_Writing.ipynb
│   └── README.md
├── 4 Tool Use and Conversational Chess
│   ├── L4_Tool_Use_and_Conversational_Chess.ipynb
│   └── README.md
├── 5 Coding and Financial Analysis
│   ├── L5_Coding_and_Financial_Analysis.ipynb
│   └── README.md
├── 6 Planning and Stock Report Generation
│   ├── L6-Planning_and_Stock_Report_Generation.ipynb
│   └── README.md
├── AI_agent_autogen.py
├── AI_agents_thesis
│   └── README.md
├── OpenAI_API
│   ├── GPT3-5-turbo.py
│   ├── GPT4.py
│   ├── README.md
│   ├── freecodecamp.py
│   ├── main.py
│   └── short_story.txt
├── README.md
├── coding
├── finetune_LLM
│   ├── README.md
│   └── sentiment_analysis.py
├── hugging-face
│   ├── GPT2.py
│   ├── GPTNeo.py
│   └── README.md
├── img
│   ├── conversations.png
│   ├── customer_onboarding_task.png
│   ├── intro.png
│   ├── nested chats.png
│   └── reflection.png
└── requirements.txt

Getting started

pip install pyautogen

Example of code

• define agents
• define chat
• initiate chat
• see the chat history, and also the cost

from autogen import ConversableAgent

llm_config={"model":"gpt-3.5-turbo"}

agent = ConversableAgent(
    name='chatbot',
    system_message='You are a chatbot and you are an expert financial advisor'
    llm_config=llm_config,
    human_input_mode='NEVER'
)

By setting the system message, we can define the behavior of the agent.

Summary methods:

• "reflection_with_llm"

summary_prompt, to instruct the llm on how to do the summary

References

1. AI Agentic Design Patterns with AutoGen, DeepLearning.AI, short course
2. AutoGen website, an Open-Source Programming Framework for Agentic AI
   1. Getting started