Constructing an AI Agent Tutorial – Half 1

The usage of the time period “AI Agent” has elevated by 10x within the final 1 yr, as per knowledge from Google Developments. This surge displays a broader shift: folks and organizations more and more need AI Techniques that not solely reply questions, but additionally take actions on their behalf. From simplifying mundane duties to streamlining enterprise operations, the promise of Agentic AI is capturing international consideration.

So, what does this actually imply in observe? Let’s begin with a relatable state of affairs of how AI Brokers can rework on a regular basis duties within the close to future. Think about planning for a trip, which entails reserving motels, flights, and rental automobiles. Immediately, this course of is fragmented and time-consuming. In an Agentic AI world, nevertheless, we may merely present a easy immediate that may generate tailor-made journey packages, full with itineraries, eating places, and bookings.

Right here is an instance immediate that might work in such a state of affairs:

“I wish to guide a household journey with 2 youngsters within the months of June/July for a weekend plus 2 days. Don’t embrace the 2nd week and third week of June. I’d simply want to hold two cabin luggage, and like tasting the perfect native meals. Plan for an itinerary not longer than 2-3 hours drive from the town.”

On this article, we’ll transcend the buzzword that’s AI Brokers. You’ll first perceive the basics of AI Brokers after which discover the platforms that make them attainable. Lastly, we’ll construct a hands-on mission: a YouTube Summarizer Agent utilizing the Phidata framework. By the tip, you’ll know what Agentic AI is and tips on how to begin constructing one with the SOTA instruments.

Be aware: That is the primary article in a two-part collection on constructing AI Brokers from the bottom up. On this article, we’ll discover the worth of AI Brokers, introduce standard Agentic AI platforms, and stroll by way of a hands-on tutorial for constructing a easy AI Agent. The subsequent a part of the collection will dive deeper with a hands-on tutorial. There, we’ll construct Brokers that may automate duties and work together with exterior instruments and APIs.

Fundamentals of AI Brokers

In easy phrases, AI Brokers are programs that may carry out duties autonomously by deciphering the info from the setting. AI brokers could make selections based mostly on that knowledge to realize the objectives. Consider them as orchestrators, connecting varied instruments, utilizing Giant Language Fashions (LLM) to motive, plan, and execute duties. For an in depth introduction to LLMs, you possibly can check with this text.

Let’s break down this definition utilizing the above trip planning instance:

• Carry out duties autonomously: Ebook flight, lodge, and rental automobile reservations by way of the respective distributors.
• Deciphering the info: Account for components like climate, visitors, and native occasions to recommend the perfect actions that match the tempo.
• Making selections: Think about there are dozens of eating places obtainable, Brokers can present suggestions based mostly on the indicated desire and previous opinions.
• Obtain objectives: Put collectively a journey plan that matches the necessities – dates, period, preferences, and household wants.

Agentic AI Platforms

An Agentic AI framework is a toolkit that allows the creation of AI programs able to reasoning, planning, and taking actions autonomously or semi-autonomously by way of software use and reminiscence. In brief, these frameworks present the construction wanted to create brokers.

There are a number of standard Agentic AI platforms, similar to LangChain, CrewAI, and Phidata. For this tutorial, we’ll use Phidata – a light-weight and developer-friendly platform. Phidata comes with built-in entry to a wide range of instruments and LLMs. This permits us to construct and deploy AI Brokers inside only a few traces of code.

Construct a YouTube summarizer Agent

The YouTube Summarizer Agent is designed to extract key insights and details from any YouTube video. It saves time by offering concise summaries without having to look at your entire content material. For the aim of the tutorial, we’ll use Google Colab pocket book to write down and execute the code and Phidata Agentic AI Platform to energy the Agent.

Mannequin: Inside Phidata, we’ll leverage the Groq mannequin internet hosting platform. It’s an inference service that runs LLMs on a devoted GPU infrastructure. Be aware that it’s totally different from Grok, which is an LLM from xAI. Since LLMs are resource-intensive, utilizing Groq helps to dump computation from the native {hardware} or Colab-provided {hardware}. This ensures quicker and extra environment friendly execution. Groq has entry to a number of fashions from totally different LLM suppliers. (see full checklist right here)

Instruments: To retrieve YouTube video knowledge, we’ll use the built-in Device from the Phidata framework (known as YouTube Instruments). This software helps us entry video metadata and captions. The agent then passes these to the chosen LLM to generate correct and insightful summaries.

Right here is the code for a YouTube summarizer agent:

from phi.agent import Agent
from phi.mannequin.groq import Groq
from phi.mannequin.openai import OpenAIChat
from phi.instruments.youtube_tools import YouTubeTools


agent = Agent(
    # mannequin=Groq(id="llama3-8b-8192"),
    mannequin=Groq(id="llama-3.3-70b-versatile"),  ## Toggle with totally different LLM mannequin
    instruments=[YouTubeTools()],
    show_tool_calls=True,
    # debug_mode=True,
    description="You're a YouTube agent. Receive the captions of a YouTube video and reply questions.",
)


agent.print_response("Summarize this video https://www.youtube.com/watch?v=vStJoetOxJg", markdown=True, stream=True)

Following is the output generated by the YouTube Summarizer agent (above code). The YouTube hyperlink within the above code is a video of Andrew Ng on the Machine Studying specialization. As proven beneath, it precisely summarizes the video content material. Be aware that the response could differ for every run due to the probabilistic nature of LLMs.

Detailed Tutorial

Listed below are the step-by-step directions for creating the YouTube Summarizer agent.

1. Clone Pocket book

• Clone Colab pocket book right here (it requires a Google account)
• Set up dependencies (first cell with code)

2. Get API key for Groq

With the intention to run the Agent, on condition that we use the Groq mannequin internet hosting platform, we want an account with Groq. Comply with the steps beneath to enroll / log in to Groq and get an API key.

– Go to the Groq Developer Portal: Open your browser and go to: https://console.groq.com

– Signal Up or Log In

• If you have already got an account, click on Log In.
• In the event you’re new, click on Signal Up and observe the prompts to create an account (it’s possible you’ll have to confirm your electronic mail).

– Entry the API Part

• As soon as logged in, you’ll land on the Groq Console.
• Navigate to the API Keys part from the sidebar or dashboard.

– Generate a New API Key

• Click on the “Create API Key” button.
• Give your key a reputation (e.g., “workshop-key”).
• Click on Create or Generate.

– Copy and Retailer the Key Securely

• Your API key will likely be proven solely as soon as — copy it instantly and retailer it in a safe location.
• By no means expose your API key in client-side code or public repositories.

3. Add the API key within the Secret Supervisor

• Click on on Secrets and techniques (Key signal) on the left pane of Colab
• Present the title as GROQ_API_KEY and the Worth because the API Key copied in Step 5 above
• Toggle “ON” the pocket book entry.

Conclusion

On this article, we explored the rising demand for an AI Agent and walked by way of a real-world instance of how they will simplify on a regular basis duties. We broke down the basics of AI Brokers and a few standard Agentic AI Frameworks. We additionally constructed a hands-on mission: a YouTube Summarizer Agent powered by Phidata.

That is just the start. Within the second article of this collection, we’ll go deeper by constructing a research planner agent that doesn’t simply generate plans but additionally takes actions. It would create duties in Jira, ship calendar invitations, and show how AI Brokers can seamlessly combine with exterior instruments and APIs to automate real-world workflows.

Take a look at the half 2 of this collection right here – Constructing Research Planner Agent: AI Agent Tutorial Half 2

Co-Creator for the article: Abhishek Agrawal

Praveen is a seasoned Information Scientist, with over a decade of expertise in analytics. He has tackled complicated enterprise challenges and pushed innovation by way of data-driven choice making. His experience spans throughout areas similar to Machine Studying, Statistics, and Scalable Analytics, serving to to launch a number of revolutionary merchandise.