In the second part of the tutorial series on AI Agents, we explore how to improve their reliability using iterations and chains. AI Agents can perform tasks by using various tools, but they often need to refine their strategies when faced with challenges. This tutorial provides Python code for setting up an Agent that can gather financial data and how it can retry actions if the first attempt doesn’t succeed. By creating a structured approach to problem-solving, the Agent learns to use different tools effectively. The article ends with an invitation to look forward to more advanced examples in Part 3 and includes a link to the full code on GitHub.
Exploring AI Agents: Iterations and Chains for Enhanced Performance
In the first part of our series on AI Agents, we introduced these autonomous programs that perform tasks and make decisions. Now, we’ll dive deeper into improving their reliability and effectiveness through iterations and chains.
AI Agents utilize tools to accomplish tasks, but they often face challenges when tools don’t work on the first try. To address this, Agents need to organize tasks logically and adapt their strategies based on the situation. This leads us to two main methods for enhancing their functionality:
- Iterations: The process of repeating an action to refine it with each attempt, helping the Agent reach optimal solutions.
- Chains: A sequence of actions where the result of one step influences the next, ensuring a smooth workflow.
In this tutorial, we will show you how to implement these strategies using Python code that you can easily apply to similar projects. We will use the Yahoo Finance API to gather financial data.
Setting Up Your Environment
Here’s a simple snippet to get you started:
import Ollama
llm = "qwen2.5"Next, we’ll download Microsoft stock data for the last five days using the Yahoo Finance API:
import yfinance as yf
stock = "MSFT"
yf.Ticker(ticker=stock).history(period='5d')Now, let’s turn this into a useful tool that the AI Agent can use.
Introducing Iterations
Iterations allow an Agent to repeat an action until it succeeds or a specified number of attempts have been reached. This way, if the initial inputs aren’t effective, the Agent can try different options based on previous attempts.
max_i, i = 3, 0
while res == 'no' and i < max_i:
# Logic for retries with updated feedback
If the Agent tries three times and still fails, it will inform the user but this time will be more informed with prior attempts documented. This built-in feedback improves the likelihood of success.
Leveraging Chains for Improved Functionality
A chain connects steps linearly, using the output of one as the input for the next. In our example, we can add a web search tool to help the Agent retrieve information if the stock data tool fails.
from langchain_community.tools import DuckDuckGoSearchResults
def search_web(query:str) -> str:
return DuckDuckGoSearchResults(backend="news").run(query)By setting up a structured prompt, we can effectively guide the Agent in using both tools to resolve queries about stock data. The prompt should clearly define the Agent's goals and the warnings it should consider.
prompt = '''
[GOAL] Assist the user using available tools.
[RETURN] Stock data requested by the user.
[WARNINGS] The correct ticker symbol is essential for retrieving data.
[CONTEXT] First, attempt 'get_stock'. If it fails, use 'search_web' to find information and then retry 'get_stock' with the correct ticker symbol.
'''This structured approach sets up the Agent to know when and how to utilize its tools effectively. If the first attempt fails, the Agent can quickly pivot to the second tool and learn from the information it gathers.
Conclusion
In this blog post, we explored how to enhance AI Agents' reliability through iterations and chains. By incorporating these techniques, you're well-equipped to create robust Agents for a variety of applications. Keep an eye out for part three, where we will delve into even more advanced examples.
If you're interested in the full code from this article, you can find it on GitHub. Feel free to reach out with questions or share your own projects.
Let’s connect and learn together!
What is AI Agents from Zero to Hero — Part 2?
AI Agents from Zero to Hero — Part 2 is a course designed to help learners understand and create their AI agents. It builds on basic knowledge and dives deeper into advanced concepts, making it easier for anyone to learn and apply AI in real-world situations.
Who is this course for?
This course is for anyone interested in AI, whether you are a beginner or someone with some experience. If you want to learn how to develop AI agents and improve your skills, this course is perfect for you.
What will I learn in Part 2?
In Part 2, you will learn about advanced techniques in AI agent development. This includes topics like machine learning, natural language processing, and how to integrate AI agents into different applications. You will also get hands-on experience through practical projects.
Do I need to know programming to take this course?
Having some basic programming knowledge can be helpful, but it's not required. The course is structured in a way that even those with limited or no coding experience can follow along and learn effectively.
How long does it take to complete the course?
The length of the course can vary depending on your pace, but most learners complete it in a few weeks. You can take your time to absorb the material, complete exercises, and work on projects at your own speed.