Teamwork Makes the Dream Work - Build Agentic Workflow
Build an agentic workflow that analyzes Reddit posts and generates a report based on the analysis. All using only local models.
Agentic workflows automate repetitive tasks through AI agents that follow predefined steps while using provided tools. In this tutorial, you'll create AgentGist, a local system that analyzes Reddit posts and generates reports based on the analysis.
What makes this project special? It runs entirely on your machine using Ollama, adapts during execution, and lets you filter posts with semantic search based on your interests.
Tutorial Goals
- Create your first agentic workflow using LangGraph and Pydantic
- Use a human-in-the-loop approach to alter the workflow dynamically
- Create custom tool to enhance your agent's abilities (filter posts)
- Practice collecting real data from Reddit
- Build UI to interact with your workflow and present the output
Do You Need A Library to Build Agents?
The short answer is no. You can build agentic applications with just LLM calls, tools, and some loops. However, writing this from scratch can be challenging, especially when you're trying to build and ship quickly.
This is where LangGraph1 comes in:
LangGraph is a library for building stateful, multi-actor applications with LLMs, used to create agent and multi-agent workflows.
Created by the LangChain team, LangGraph lets you define workflows as state graphs with added tools and utilities. For a deeper dive, check their documentation: LangGraph Quickstart.
For this tutorial, we'll use LangGraph's Functional API2, which simplifies workflow creation with decorated Python functions. We'll also use Pydantic3 to define data models, structure our agent outputs, and validate data—building on what you've learned in previous tutorials.