Agentic AI Design Patterns

Welcome to the Agentic AI Design Patterns repository. This project showcases various "Agentic AI" design patterns popularized by DeepLearning.AI, featuring functional demos of autonomous, iterative, and collaborative agentic AI applications. The code in this repository was developed with the assistance of Google AntiGravity.

Instead of expecting an LLM to generate a perfect result in a single pass, these agentic patterns enable models to reason, utilize tools, critique their own outputs, and facilitate collaboration among specialized agents.

This repository demonstrates the following core design patterns:

1. Reflection
2. Tool Use (Function Calling) (integrated into the demos below)
3. Reason and Act (ReAct)
4. Planning
5. Multi-Agent Collaboration

Getting Started

Prerequisites

• Python 3.11+
• Docker
• Ollama (Required for local LLM inference)

Installation

1. Clone the repository:

   git clone https://github.com/shomec/agentic-ai-designpatterns.git
   cd agentic-ai-designpatterns

2. Run the Demos:

   To explore a pattern, navigate to its directory and use Docker Compose to build and run the application.

   • Reflection Pattern

     cd reflection
     docker compose up --build

   • Reason and Act (ReAct) Pattern

     cd reason-and-act
     docker compose up --build

   • Planning Pattern

     cd planning
     docker compose up --build

   • Multi-Agent Collaboration Pattern

     cd multi-agent
     docker compose up --build

   Note: The Tool Use pattern is a fundamental capability demonstrated within the ReAct, Planning, and Multi-Agent demos.

Tech Stack

This project leverages the following technologies to implement these patterns:

• Ollama: Orchestrates local Large Language Models (specifically gemma3:1b).
• Model Context Protocol (MCP): Provides standardized, containerized access to external tools (e.g., DuckDuckGo Search) within the Docker environment.
• Docker: Ensures consistent and isolated execution environments for all services.