Create Basic Visualizations

A picture is worth a thousand rows. This tutorial shows how the agent interprets a plain-English request, selects appropriate chart types based on data characteristics, and generates publication-ready PNG files — no matplotlib boilerplate required.

You'll see the agent create three charts (a box plot, a bar chart, and a violin plot) from a single prompt, with reasoning shown at every decision point.

Setup

npx @votee-ai/magic-data-agent-skills init

mkdir magic-tutorial && cd magic-tutorial

curl -O https://docs.votee.ai/tutorials/data/product_metrics.csv

Sample Data

This tutorial uses a 300-row product metrics dataset with clean numeric and categorical columns. Download it to follow along:

The Agent Trace

The trace below shows how the agent reads the data, selects chart types based on column characteristics, writes a generation script, and verifies each chart visually.

{
  "file_path": "data/product_metrics.csv"
}

{
  "rows": 300,
  "columns": 9,
  "categories": 8,
  "records_per_category": "29–54",
  "numeric_columns": [
    "sales",
    "profit_margin",
    "units_sold",
    "customer_rating",
    "return_rate"
  ]
}

import pandas as pd
df = pd.read_csv('data/product_metrics.csv')
print(df['category'].value_counts())
print(df.describe())

{
  "charts": [
    {
      "type": "box plot",
      "metric": "sales",
      "groupby": "category"
    },
    {
      "type": "box plot",
      "metric": "profit_margin",
      "groupby": "category"
    },
    {
      "type": "box plot",
      "metric": "customer_rating",
      "groupby": "category"
    },
    {
      "type": "dashboard",
      "metrics": [
        "sales",
        "profit_margin",
        "customer_rating"
      ]
    }
  ],
  "output_dir": "charts/",
  "dpi": 300,
  "palette": "Okabe-Ito (colorblind-safe)"
}

{
  "files_created": 4,
  "sales_by_category.png": "Box plot — sales distribution sorted by median, Y-axis floored at 0",
  "profit_margin_by_category.png": "Box plot — profit margin distribution",
  "customer_rating_by_category.png": "Box plot — customer rating distribution (1-5 scale)",
  "dashboard_by_category.png": "All three side-by-side for comparison"
}

import pandas as pd
import matplotlib.pyplot as plt
from pathlib import Path

df = pd.read_csv('data/product_metrics.csv')
OKABE_ITO = ['#E69F00','#56B4E9','#009E73',
  '#F0E442','#0072B2','#D55E00','#CC79A7','#000000']

def box_by_category(df, col, ylabel, title, path, y_floor_zero=False):
    cats = df.groupby('category')[col].median().sort_values().index
    data = [df[df['category']==c][col] for c in cats]
    bp = ax.boxplot(data, labels=cats, patch_artist=True)
    for patch, color in zip(bp['boxes'], OKABE_ITO):
        patch.set_facecolor(color)

{
  "directory": "charts/"
}

{
  "files_verified": 4,
  "resolution": "2970×1770 px (300 DPI)",
  "metadata_files": 3,
  "all_valid": true
}

ls -la charts/

Data Flow

Charts Generated

The agent produced three charts — click to view full size:

What Happened

Try It Yourself

Copy this prompt with your own CSV:

Create visualizations showing the distribution of sales, profit margins,
and customer ratings by category from data/product_metrics.csv.
Save charts to a charts/ directory.

Want to be more specific? Try targeting individual metrics:

Show me a box plot of revenue by region and a scatter plot of quantity vs price

The agent accepts both broad intent ("visualizations") and precise column-level instructions.

What You Learned

• Intent mapping: "distribution" → box plots; "by category" → grouped comparisons. You don't need to name chart types explicitly.
• The agent writes a reusable script (generate_charts.py) rather than throwaway commands, so you can regenerate or modify charts later.
• Visual verification is built in — the agent reads each PNG to confirm the chart rendered correctly before reporting success.
• Multiple charts are produced from one prompt, with consistent styling across all outputs.

Alternate Approaches

Next Steps