CrewAI: Agentic Exploratory Data Analysis

This notebook demonstrates a real-world CrewAI application where a Data Analyst agent performs Exploratory Data Analysis (EDA) on a CSV dataset using three custom tools.

Workflow¶

Generate a synthetic sales dataset
Define three EDA tools (preview, histogram, grouped aggregation)
Create a Data Analyst agent equipped with those tools
Assign a multi-step EDA task and execute the crew
Display the histogram artifact produced by the agent

All reusable helper functions live in crewai_utils.py; this notebook only orchestrates and displays results.

%load_ext autoreload
%autoreload 2

import logging

import matplotlib.pyplot as plt
import matplotlib.image as mpimg

from crewai import Agent, Crew, Process, Task

import crewai_utils as tcrwuti

logging.basicConfig(level=logging.INFO)
_LOG = logging.getLogger(__name__)

Part 1: Data Preparation¶

Objective: Create a synthetic sales dataset to use throughout the tutorial.

Columns: region, month, units_sold, price
20 rows; random seed fixed at 42 for reproducibility
Saved to data/sales.csv

# Generate (or regenerate) the demo sales CSV.
csv_path = tcrwuti.generate_sales_csv("data/sales.csv")
print(f"Dataset saved to: {csv_path}")

# Quick sanity check – preview the first few rows.
import pandas as pd

df = pd.read_csv(csv_path)
print(df.head())

Part 2: LLM and Agent Setup¶

Objective: Configure a local Ollama LLM and create a Data Analyst agent with three EDA tools:

read_head – preview CSV rows
plot_histogram – generate a histogram PNG
groupby_agg – compute grouped means

All tools are defined in crewai_utils.EDA_TOOLS.

# Use the local Ollama LLM so no API key is required.
llm = tcrwuti.get_local_llm(
    model="ollama/gemma3:latest",
    base_url="http://host.docker.internal:11434",
    temperature=0,
)

# Create the Data Analyst agent.
analyst = Agent(
    role="Data Analyst",
    goal=(
        "Perform lightweight EDA on local CSVs via tools and "
        "report succinct results."
    ),
    backstory=(
        "Prefers precise, minimal outputs. Uses tools exactly as requested."
    ),
    tools=tcrwuti.EDA_TOOLS,
    llm=llm,
    verbose=True,
)
print(f"Agent tools: {[t.name for t in analyst.tools]}")

Part 3: EDA Task Definition¶

Objective: Define a three-step EDA task for the agent.

The task instructs the agent to:

Preview the top 5 rows of the CSV
Plot a histogram of the price column
Compute mean units_sold grouped by region

The expected_output anchors the agent’s final response format.

eda_task = Task(
    description=(
        "1) Use read_head on './data/sales.csv' (n=5).\n"
        "2) Use plot_histogram on column 'price' with 20 bins.\n"
        "3) Use groupby_agg by 'region' on metric 'units_sold'.\n"
        "Return three sections: PREVIEW, HISTOGRAM_PATH, GROUPED_MEANS, "
        "each containing only the respective tool output."
    ),
    expected_output="Sections: PREVIEW, HISTOGRAM_PATH, GROUPED_MEANS.",
    agent=analyst,
)

Part 4: Crew Execution¶

Objective: Assemble the crew and run the EDA pipeline.

We use Process.sequential since there is only a single task.

crew = Crew(
    agents=[analyst],
    tasks=[eda_task],
    process=Process.sequential,
    verbose=True,
)

# Run the crew and capture the structured result.
result = crew.kickoff()
print("\n=== RESULT ===\n", result)

Part 5: Display Artifacts¶

Objective: Show the histogram PNG generated by the agent.

# Display the histogram generated by the plot_histogram tool.
hist_path = "artifacts/hist_price.png"
try:
    img = mpimg.imread(hist_path)
    fig, ax = plt.subplots(figsize=(7, 4))
    ax.imshow(img)
    ax.axis("off")
    ax.set_title("Histogram of price (generated by CrewAI agent)")
    plt.tight_layout()
    plt.show()
except FileNotFoundError:
    print(f"Histogram not found at {hist_path}. Run the crew first.")