Agents at Work— Navigating The Digital Realm With Langchain

Greetings and welcome to XenAIBlog! It's been a whole year since the birth of ChatGPT—Happy Birthday indeed! Time sure does fly.

On that note, let's get down to business. What exactly are Agents in Langchain? They take center stage as the brains behind our digital interactions, bringing a fresh perspective to how we engage with technology. Unlike the old-fashioned world of code, these agents are dynamic thinkers, relying on language models (LLMs) to analyze what users throw their way and figure out the best course of action.

Agents are like sharp problem-solvers, equipped with a toolkit featuring search engines, databases, and websites—think Sheldon Cooper from "Young Sheldon." The real powerhouse, much like the genius behind the scenes in the show, is the LLM. It takes in what users ask for and maps out a series of actions to get things done, just like Sheldon. Agents, grouped into categories, use these language models for all kinds of tasks—answering questions, tackling tricky problems, and pulling in info from various sources, much like our favorite young genius.

The key to their success lies in the arsenal of tools tucked away in their back pockets—functions designed to perform specific tasks. LangChain steps in as the helping hand, providing these agents with a set of built-in tools. It's all about ensuring these agents are well-equipped and prepared for any task that comes their way.

We explored two types of Agents in our projects: Action Agents and Plan-and- Execute Agents. Action Agents are the doers. They make decisions step by step, like that friend who tackles problems head-on. On the other hand, Plan-and- Execute Agents are like the strategic masterminds. They plan ahead for the big stuff, like a friend who's always three steps ahead in a chess game. This method, known as Plan-and-Solve prompting, shows how good they are at thinking ahead and being organized. It's like having a game plan for whatever comes their way.

Now, how did we use these agents to scour through the web? LangChain hooks agents up with some heavy-duty tools like SerpAPI and llm-math. SerpAPI is a search pro—it scrapes data from all kinds of search engines and results pages. Then there's llm-math, the math guru, using big language models for some brainy calculations. Agents become the internet navigators, smoothly cruising around, grabbing info, and flexing their math muscles when needed. Easy peasy!

We'll now show you some behind-the-scenes action with some of these agents. We first started with installing the required packages.

Next, we went on to import the necessary libraries:

Then, we had to define the environment variables. Do take note that in this example, we are using the Azure OpenAI API. utilizing the Azure OpenAI API. To create an Azure OpenAI resource, refer here.

Moving forward, we set up the API key for SerpAPI. SerpAPI is basically a commercial service that provides access to Google Search results. Now, that's something!

To maintain security and fair resource usage, we used an API key. You can generate your API key by registering right here.

Remember, llm-math is an integral tool within the LangChain framework and so, we did not require a key for it.

After successfully configuring all the environment variables, we initialized the Azure OpenAI model with a specific API key and temperature. This initialization process also involved loading tools for both search and math, ultimately leading to the creation of an agent.

The template utilized by the agent for generating prompts looked like this:

Next, it was time for some action! We asked the agent a question based on the latest news:

Look at that! We got an accurate and up-to-date response. Then, we posed a math problem:

Impressive! The model pulled in info from the web, did some slick calculations, and boom – precise results delivered like it was no big deal.

Agents are like well-oiled machines, ready to tackle anything that comes its way, seamlessly handling tasks ranging from staying informed with the latest news to performing complex mathematical computations.

Hope you all learned something cool today!

Adiós, until next Friday!