AI Agents 💫

A curated collection of Jupyter notebooks exploring design patterns for autonomous AI agents using frameworks like Microsoft AutoGen, LangGraph, Taskweaver, CrewAI, and more.

---

📌 Overview

This repository demonstrates how to build, orchestrate, and compose intelligent agents capable of planning, reasoning, code execution, retrieval-augmented generation (RAG), and multi-agent collaboration. Each notebook is self-contained and focused on a specific pattern or use case.

---

🌟 Agent Architecture

---

📁 Notebooks

1. 🔬 Multi-Agent Research Group Chat

File: agentchat_groupchat_research.ipynb

A multi-agent group chat where specialized agents — Admin, Engineer, Scientist, Planner, Executor, and Critic — collaborate to perform research tasks. Demonstrates the impact of a Critic agent on output quality by comparing results with and without it.

• Task: Find recent LLM papers on arXiv and organize them into a markdown table by domain
• Framework: AutoGen
• Key agents: UserProxyAgent, AssistantAgent, GroupChatManager

---

2. 📊 Data Analysis with Visualization Critic

File: Autogen_Data_Analysis_Graphs.ipynb

A group chat between a Coder agent and a Visualization Critic agent. The critic reviews generated plots and suggests improvements, which the coder implements — producing higher-quality visualizations iteratively.

• Tasks: Plot weight vs. horsepower from car data; visualize Seattle weather frequency
• Framework: AutoGen
• Libraries: matplotlib, seaborn, pandas
• Key agents: UserProxyAgent, Coder AssistantAgent, Critic AssistantAgent

---

3. 📚 Group Chat with Retrieval-Augmented Generation (RAG)

File: Autogen_groupchat_RAG.ipynb

Demonstrates how RetrieveUserProxyAgent overcomes the knowledge limitations of standard LLMs by grounding responses in provided documentation. Compares standard chat (incorrect output) vs. RAG-enabled chat (correct output) for tasks involving post-training API knowledge.

• Framework: AutoGen + ChromaDB
• Use case: Generate valid Spark API code using FLAML documentation
• Key agents: RetrieveUserProxyAgent, Boss, Boss_Assistant

---

4. 🗺️ Collaborative Planning with a Planner Agent

File: Autogen_planning_with_Agents.ipynb

Shows how an Assistant agent can consult a dedicated Planner agent before executing tasks. The Planner breaks down complex tasks into actionable steps without writing code itself, while the Assistant handles execution.

• Task: Suggest a fix for an open "good first issue" on the FLAML GitHub repo
• Framework: AutoGen
• Pattern: ask_planner function call between agents
• Planner component:

---

5. 🦙 Local RAG Agent with LLaMA 3

File: Local_RAG_Agent_LLAMA_3.ipynb

An advanced, fully local RAG agent built with LangGraph that combines three research-backed RAG strategies:

Strategy	Paper	Description
Adaptive RAG	arXiv 2403.14403	Routes questions to vectorstore or web search
Corrective RAG	arXiv 2401.15884	Falls back to web search when docs are irrelevant
Self-RAG	arXiv 2310.11511	Self-corrects hallucinations and incomplete answers

• Framework: LangGraph + LangChain
• LLM: LLaMA 3 (via Ollama — runs fully locally)
• Embeddings: GPT4All
• Vector store: Chroma
• Web search: Tavily
• Pipeline stages: Index → Retrieve → Grade → Generate → Hallucination check → Answer grading

---

🏗️ Agent Design Patterns Covered

Pattern	Description
Multi-Agent Group Chat	Multiple specialized agents collaborate in a shared conversation
Plan & Execute	A Planner decomposes tasks; an Executor implements them
RAG with Agents	Agents retrieve and ground responses in external documents
Critic/Reflection	A Critic agent reviews outputs and triggers self-improvement
Adaptive Routing	An agent routes queries to the best retrieval strategy
Self-Correction	Agents detect and fix hallucinations in their own outputs

---

⚙️ Setup & Installation

Prerequisites

• Python 3.9+
• OpenAI API key (for AutoGen notebooks) or Ollama with LLaMA 3 (for local RAG notebook)

AutoGen Notebooks

pip install pyautogen
pip install pyautogen[retrievechat]  # for RAG notebook

Configure your API key in OAI_CONFIG_LIST:

[
  {
    "model": "gpt-4",
    "api_key": "<your OpenAI API key>"
  }
]

Local RAG Notebook (LLaMA 3)

pip install langchain-nomic langchain_community tiktoken langchainhub \
            chromadb langchain langgraph tavily-python

Install and run Ollama with LLaMA 3:

ollama pull llama3

📚 References & Further Reading

• AutoGen Documentation
• LangGraph Documentation
• Taskweaver by Microsoft
• Adaptive RAG Paper
• Corrective RAG Paper
• Self-RAG Paper

---

---

⭐ If you find this repository helpful, please consider giving it a star!