1. Project Scoping: Search Bot
Defining the MVP. We will build an agent that takes a topic, searches the web, reads pages, and writes a comprehensive summary report automatically.
It is finally time to write some code. In this module, we will build a complete, end-to-end autonomous agent. The biggest mistake beginners make is trying to build a general-purpose “AGI” that can do everything. That will fail.
Instead, we will build a highly constrained, highly useful Minimum Viable Product (MVP): The Autonomous Research Bot.
The Goal
We want to be able to type a single command into our terminal:
python run_agent.py "What are the latest developments in solid-state batteries?"
The agent should then wake up, realize it needs to search the internet, execute a Google search, click on the top 3 links, read the text on those websites, synthesize the information into a 500-word markdown report, and save that report to our hard drive.
🎯 Defining the Boundaries
- LLM Engine: We will use
Llama-3-8B-Instructrunning locally via Ollama to keep things free and private. - Framework: We will use a lightweight custom ReAct loop instead of a heavy framework so you can see exactly how the mechanics work under the hood.
- Tools: The agent will have exactly three tools:
search_web,read_url, andsave_file.
Why this specific project?
This project touches every fundamental pillar of agent development.
- It requires the LLM to make a plan (Search first, Read second, Save third).
- It requires the LLM to format JSON correctly to call the external API (Google Search).
- It requires the system to manage a context window (HTML from websites is massive, so we have to parse it before feeding it to the LLM).
- It requires the agent to handle errors (What if a website blocks the scraper? The agent must realize this and try the next link).