Build RAG-Powered AI Agents on Custom Data in Google Colab

---

---

AI Summary

Video Summary: Building a RAG-Powered AI Agent with Llama Index on Google Colab

• Objective: Demonstrate how to create a RAG-powered AI agent using Llama Index and Crew AI on Google Colab for free.
• RAG-Powered AI Agent:
  • RAG stands for Retrieval Augmented Generation.
  • Integrates external data with AI agents to generate context-specific information.
• Tools Used:
  • Llama Index: A framework for building large language model applications with custom data.
  • Crew AI: A framework for creating customizable AI agents for specific tasks.
• Integration:
  • Various models, including API-based ones like OpenAI Gemini, can be integrated with Crew AI.
  • Open-source software like AMA can also be used locally.
• Example Data:
  • A fictitious finance CSV file with product sales data by region, country, and year.
• Google Colab Setup:
  • Installation of Llama Index and integrations.
  • Importing necessary libraries and uploading the finance CSV file.
• Data Loading and Indexing:
  • Using Llama Index’s directory reader to load the CSV file.
  • Creating a vector store and index for the data.
• Large Language Model Setup:
  • Using OpenAI’s GPT-4 model.
  • Setting up API keys and environment variables in Google Colab.
• Building the AI Agent:
  • Defining a “researcher” agent as a senior market analyst.
  • Defining a “writer” agent to produce content based on research insights.
  • Creating tasks for data analysis and blog post writing.
  • Forming a Crew AI team with the researcher and writer agents.
• Execution and Results:
  • Kicking off the process with crew.kick_off.
  • Agents collaborate to analyze data and generate a blog post.
  • The process results in a markdown-formatted blog post.
• Conclusion:
  • The video showcases the integration of Crew AI with custom data and Llama Index to build a RAG-powered application.
  • The code will be shared in the video description for viewers to experiment with.
  • Viewers are encouraged to subscribe and share the content.

Link to the code in the video description (Note: This is a placeholder as the actual link is not provided in the summary).

!pip install llama-index-core

!pip install llama-index-readers-file

!pip install llama-index-embeddings-openai

!pip install llama-index-llms-llama-api

!pip install ‘crewai[tools]‘

import os

from crewai import Agent, Task, Crew, Process

from crewai_tools import LlamaIndexTool

from llama_index.core import SimpleDirectoryReader, VectorStoreIndex

from llama_index.llms.openai import OpenAI

reader = SimpleDirectoryReader(input_files=[“finance.csv”])

docs = reader.load_data()

docs[1].get_content()

from google.colab import userdata

os.environ[‘OPENAI_API_KEY’]=userdata.get(‘OPENAI_API_KEY’)

llm = OpenAI(model=“gpt-4o”)

index = VectorStoreIndex.from_documents(docs)

query_engine = index.as_query_engine(similarity_top_k=5, llm=llm)

query_tool = LlamaIndexTool.from_query_engine(

    query_engine,

    name=“Finance Query Tool”,

    description=“Use this tool to lookup the financial data of products and their sales”,

)

query_tool.args_schema.schema()

researcher = Agent(

    role=“Senior Market Analyst”,

    goal=“Uncover insights about product sales trends”,

    backstory="""You work at a market research firm.

  Your goal is to understand sales patterns across different product categories.""",

    verbose=True,

    allow_delegation=False,

    tools=[query_tool],

)

writer = Agent(

    role=“Product Content Specialist”,

    goal=“Craft compelling content on product trends”,

    backstory="""You are a renowned Content Specialist, known for your insightful and engaging articles.

  You transform complex sales data into compelling narratives.""",

    verbose=True,

    allow_delegation=False,

)

Create tasks for your agents

task1 = Task(

    description="""Analyze the sales data of top 5 products in the last quarter.""",

    expected_output=“Detailed sales report with trends and insights”,

    agent=researcher,

)

task2 = Task(

    description="""Using the insights provided, develop an engaging blog

  post that highlights the top-selling products and their market trends.

  Your post should be informative yet accessible, catering to a casual audience.

  Make it sound cool, avoid complex words.""",

    expected_output=“Full blog post of at least 4 paragraphs”,

    agent=writer,

)

Instantiate your crew with a sequential process

crew = Crew(

    agents=[researcher, writer],

    tasks=[task1, task2],

    verbose=2,  # You can set it to 1 or 2 to different logging levels

)

result = crew.kickoff()

print(”######################“)

print(result)

Written by Fahd Mirza  

Email ThisBlogThis!Share to TwitterShare to FacebookShare to Pinterest

Labels: agentic, crewai, llamaindex