LLMs + Pandas: How I Use Generative AI to Generate Pandas DataFrame Summaries

datasets and are on the lookout for fast insights with out an excessive amount of guide grind, you’ve come to the appropriate place.

In 2025, datasets typically comprise thousands and thousands of rows and a whole bunch of columns, which makes guide evaluation subsequent to inconceivable. Native Massive Language Fashions can remodel your uncooked DataFrame statistics into polished, readable studies in seconds — minutes at worst. This strategy eliminates the tedious strategy of analyzing information by hand and writing government studies, particularly if the info construction doesn’t change.

Pandas handles the heavy lifting of knowledge extraction whereas LLMs convert your technical outputs into presentable studies. You’ll nonetheless want to put in writing capabilities that pull key statistics out of your datasets, nevertheless it’s a one-time effort.

This information assumes you’ve got Ollama put in domestically. For those who don’t, you’ll be able to nonetheless use third-party LLM distributors, however I received’t clarify how to connect with their APIs.

Desk of contents:

• Dataset Introduction and Exploration
• The Boring Half: Extracting Abstract Statistics
• The Cool Half: Working with LLMs
• What You May Enhance

Dataset Introduction and Exploration

For this information, I’m utilizing the MBA admissions dataset from Kaggle. Obtain it if you wish to observe alongside.

The dataset is licensed underneath the Apache 2.0 license, which suggests you should use it freely for each private and business tasks.

To get began, you’ll want a couple of Python libraries put in in your system.

After you have all the things put in, import the mandatory libraries in a brand new script or a pocket book:

import pandas as pd
from langchain_ollama import ChatOllama
from typing import Literal

Dataset loading and preprocessing

Begin by loading the dataset with Pandas. This snippet masses the CSV file, prints fundamental details about the dataset form, and exhibits what number of lacking values exist in every column:

df = pd.read_csv("information/MBA.csv")

# Primary dataset data
print(f"Dataset form: {df.form}n")
print("Lacking worth stats:")
print(df.isnull().sum())
print("-" * 25)
df.pattern(5)

Since information cleansing isn’t the principle focus of this text, I’ll preserve the preprocessing minimal. The dataset solely has a few lacking values that want consideration:

df["race"] = df["race"].fillna("Unknown")
df["admission"] = df["admission"].fillna("Deny")

That’s it! Let’s see tips on how to go from this to a significant report subsequent.

The Boring Half: Extracting Abstract Statistics

Even with all of the advances in AI functionality and availability, you in all probability don’t need to ship your complete dataset to an LLM supplier. There are a few good the explanation why.

It might devour means too many tokens, which interprets on to increased prices. Processing massive datasets can take a very long time, particularly while you’re operating fashions domestically by yourself {hardware}. You may also be coping with delicate information that shouldn’t go away your group.

Some guide work remains to be the way in which to go.

This strategy requires you to put in writing a operate that extracts key components and statistics out of your Pandas DataFrame. You’ll have to put in writing this operate from scratch for various datasets, however the core thought transfers simply between tasks.

The get_summary_context_message() operate takes in a DataFrame and returns a formatted multi-line string with an in depth abstract. Right here’s what it contains:

• Whole utility depend and gender distribution
• Worldwide vs home applicant breakdown
• GPA and GMAT rating quartile statistics
• Admission charges by educational main (sorted by fee)
• Admission charges by work trade (high 8 industries)
• Work expertise evaluation with categorical breakdowns
• Key insights highlighting top-performing classes

Right here’s the entire supply code for the operate:

def get_summary_context_message(df: pd.DataFrame) -> str:
    """
    Generate a complete abstract report of MBA admissions dataset statistics.
    
    This operate analyzes MBA utility information to supply detailed statistics on
    applicant demographics, educational efficiency, skilled backgrounds, and
    admission charges throughout numerous classes. The abstract contains gender and
    worldwide standing distributions, GPA and GMAT rating statistics, admission
    charges by educational main and work trade, and work expertise influence evaluation.
    
    Parameters
    ----------
    df : pd.DataFrame
        DataFrame containing MBA admissions information with the next anticipated columns:
        - 'gender', 'worldwide', 'gpa', 'gmat', 'main', 'work_industry', 'work_exp', 'admission'
    
    Returns
    -------
    str
        A formatted multi-line string containing complete MBA admissions
        statistics.
    """
    # Primary utility statistics
    total_applications = len(df)

    # Gender distribution
    gender_counts = df["gender"].value_counts()
    male_count = gender_counts.get("Male", 0)
    female_count = gender_counts.get("Feminine", 0)

    # Worldwide standing
    international_count = (
        df["international"].sum()
        if df["international"].dtype == bool
        else (df["international"] == True).sum()
    )

    # GPA statistics
    gpa_data = df["gpa"].dropna()
    gpa_avg = gpa_data.imply()
    gpa_25th = gpa_data.quantile(0.25)
    gpa_50th = gpa_data.quantile(0.50)
    gpa_75th = gpa_data.quantile(0.75)

    # GMAT statistics
    gmat_data = df["gmat"].dropna()
    gmat_avg = gmat_data.imply()
    gmat_25th = gmat_data.quantile(0.25)
    gmat_50th = gmat_data.quantile(0.50)
    gmat_75th = gmat_data.quantile(0.75)

    # Main evaluation - admission charges by main
    major_stats = []
    for main in df["major"].distinctive():
        major_data = df[df["major"] == main]
        admitted = len(major_data[major_data["admission"] == "Admit"])
        complete = len(major_data)
        fee = (admitted / complete) * 100
        major_stats.append((main, admitted, complete, fee))

    # Type by admission fee (descending)
    major_stats.type(key=lambda x: x[3], reverse=True)

    # Work trade evaluation - admission charges by trade
    industry_stats = []
    for trade in df["work_industry"].distinctive():
        if pd.isna(trade):
            proceed
        industry_data = df[df["work_industry"] == trade]
        admitted = len(industry_data[industry_data["admission"] == "Admit"])
        complete = len(industry_data)
        fee = (admitted / complete) * 100
        industry_stats.append((trade, admitted, complete, fee))

    # Type by admission fee (descending)
    industry_stats.type(key=lambda x: x[3], reverse=True)

    # Work expertise evaluation
    work_exp_data = df["work_exp"].dropna()
    avg_work_exp_all = work_exp_data.imply()

    # Work expertise for admitted college students
    admitted_students = df[df["admission"] == "Admit"]
    admitted_work_exp = admitted_students["work_exp"].dropna()
    avg_work_exp_admitted = admitted_work_exp.imply()

    # Work expertise ranges evaluation
    def categorize_work_exp(exp):
        if pd.isna(exp):
            return "Unknown"
        elif exp < 2:
            return "0-1 years"
        elif exp < 4:
            return "2-3 years"
        elif exp < 6:
            return "4-5 years"
        elif exp < 8:
            return "6-7 years"
        else:
            return "8+ years"

    df["work_exp_category"] = df["work_exp"].apply(categorize_work_exp)
    work_exp_category_stats = []

    for class in ["0-1 years", "2-3 years", "4-5 years", "6-7 years", "8+ years"]:
        category_data = df[df["work_exp_category"] == class]
        if len(category_data) > 0:
            admitted = len(category_data[category_data["admission"] == "Admit"])
            complete = len(category_data)
            fee = (admitted / complete) * 100
            work_exp_category_stats.append((class, admitted, complete, fee))

    # Construct the abstract message
    abstract = f"""MBA Admissions Dataset Abstract (2025)
    
Whole Functions: {total_applications:,} individuals utilized to the MBA program.

Gender Distribution:
- Male candidates: {male_count:,} ({male_count/total_applications*100:.1f}%)
- Feminine candidates: {female_count:,} ({female_count/total_applications*100:.1f}%)

Worldwide Standing:
- Worldwide candidates: {international_count:,} ({international_count/total_applications*100:.1f}%)
- Home candidates: {total_applications-international_count:,} ({(total_applications-international_count)/total_applications*100:.1f}%)

Educational Efficiency Statistics:

GPA Statistics:
- Common GPA: {gpa_avg:.2f}
- twenty fifth percentile: {gpa_25th:.2f}
- fiftieth percentile (median): {gpa_50th:.2f}
- seventy fifth percentile: {gpa_75th:.2f}

GMAT Statistics:
- Common GMAT: {gmat_avg:.0f}
- twenty fifth percentile: {gmat_25th:.0f}
- fiftieth percentile (median): {gmat_50th:.0f}
- seventy fifth percentile: {gmat_75th:.0f}

Main Evaluation - Admission Charges by Educational Background:"""

    for main, admitted, complete, fee in major_stats:
        abstract += (
            f"n- {main}: {admitted}/{complete} admitted ({fee:.1f}% admission fee)"
        )

    abstract += (
        "nnWork Trade Evaluation - Admission Charges by Skilled Background:"
    )

    # Present high 8 industries by admission fee
    for trade, admitted, complete, fee in industry_stats[:8]:
        abstract += (
            f"n- {trade}: {admitted}/{complete} admitted ({fee:.1f}% admission fee)"
        )

    abstract += "nnWork Expertise Affect on Admissions:nnOverall Work Expertise Comparability:"
    abstract += (
        f"n- Common work expertise (all candidates): {avg_work_exp_all:.1f} years"
    )
    abstract += f"n- Common work expertise (admitted college students): {avg_work_exp_admitted:.1f} years"

    abstract += "nnAdmission Charges by Work Expertise Vary:"
    for class, admitted, complete, fee in work_exp_category_stats:
        abstract += (
            f"n- {class}: {admitted}/{complete} admitted ({fee:.1f}% admission fee)"
        )

    # Key insights
    best_major = major_stats[0]
    best_industry = industry_stats[0]

    abstract += "nnKey Insights:"
    abstract += (
        f"n- Highest admission fee by main: {best_major[0]} at {best_major[3]:.1f}%"
    )
    abstract += f"n- Highest admission fee by trade: {best_industry[0]} at {best_industry[3]:.1f}%"

    if avg_work_exp_admitted > avg_work_exp_all:
        abstract += f"n- Admitted college students have barely extra work expertise on common ({avg_work_exp_admitted:.1f} vs {avg_work_exp_all:.1f} years)"
    else:
        abstract += "n- Work expertise exhibits minimal distinction between admitted and all candidates"

    return abstract

When you’ve outlined the operate, merely name it and print the outcomes:

print(get_summary_context_message(df))

Now let’s transfer on to the enjoyable half.

The Cool Half: Working with LLMs

That is the place issues get fascinating and your guide information extraction work pays off.

Python helper operate for working with LLMs

In case you have respectable {hardware}, I strongly advocate utilizing native LLMs for easy duties like this. I exploit Ollama and the newest model of the Mistral mannequin for the precise LLM processing.

If you wish to use one thing like ChatGPT by means of OpenAI API, you’ll be able to nonetheless try this. You’ll simply want to switch the operate beneath to arrange your API key and return the suitable occasion from Langchain.

Whatever the choice you select, a name to get_llm() with a take a look at message shouldn’t return an error:

def get_llm(model_name: str = "mistral:newest") -> ChatOllama:
    """
    Create and configure a ChatOllama occasion for native LLM inference.
    
    This operate initializes a ChatOllama consumer configured to connect with a
    native Ollama server. The consumer is ready up with deterministic output
    (temperature=0) for constant responses throughout a number of calls with the
    similar enter.
    
    Parameters
    ----------
    model_name : str, optionally available
        The title of the Ollama mannequin to make use of for chat completions.
        Have to be a legitimate mannequin title that's accessible on the native Ollama
        set up. Default is "mistral:newest".
    
    Returns
    -------
    ChatOllama
        A configured ChatOllama occasion prepared for chat completions.
    """
    return ChatOllama(
        mannequin=model_name, base_url="http://localhost:11434", temperature=0
    )


print(get_llm().invoke("take a look at").content material)

Summarization immediate

That is the place you will get artistic and write ultra-specific directions on your LLM. I’ve determined to maintain issues mild for demonstration functions, however be at liberty to experiment right here.

There isn’t a single proper or improper immediate.

No matter you do, be sure to incorporate the format arguments utilizing curly brackets – these values might be crammed dynamically later:

SUMMARIZE_DATAFRAME_PROMPT = """
You're an knowledgeable information analyst and information summarizer. Your process is to absorb advanced datasets
and return user-friendly descriptions and findings.

You got this dataset:
- Identify: {dataset_name}
- Supply: {dataset_source}

This dataset was analyzed in a pipeline earlier than it was given to you.
These are the findings returned by the evaluation pipeline:


{context}


Based mostly on these findings, write an in depth report in {report_format} format.
Give the report a significant title and separate findings into sections with headings and subheadings.
Output solely the report in {report_format} and nothing else.

Report:
"""

Summarization Python operate

With the immediate and the get_llm() capabilities declared, the one factor left is to attach the dots. The get_report_summary() operate takes in arguments that can fill the format placeholders within the immediate, then invokes the LLM with that immediate to generate a report.

You may select between Markdown or HTML codecs:

def get_report_summary(
    dataset: pd.DataFrame,
    dataset_name: str,
    dataset_source: str,
    report_format: Literal["markdown", "html"] = "markdown",
) -> str:
    """
    Generate an AI-powered abstract report from a pandas DataFrame.
    
    This operate analyzes a dataset and generates a complete abstract report
    utilizing a big language mannequin (LLM). It first extracts statistical context
    from the dataset, then makes use of an LLM to create a human-readable report within the
    specified format.
    
    Parameters
    ----------
    dataset : pd.DataFrame
        The pandas DataFrame to research and summarize.
    dataset_name : str
        A descriptive title for the dataset that might be included within the
        generated report for context and identification.
    dataset_source : str
        Details about the supply or origin of the dataset.
    report_format : {"markdown", "html"}, optionally available
        The specified output format for the generated report. Choices are:
        - "markdown" : Generate report in Markdown format (default)
        - "html" : Generate report in HTML format
    
    Returns
    -------
    str
        A formatted abstract report.
    
    """
    context_message = get_summary_context_message(df=dataset)
    immediate = SUMMARIZE_DATAFRAME_PROMPT.format(
        dataset_name=dataset_name,
        dataset_source=dataset_source,
        context=context_message,
        report_format=report_format,
    )
    return get_llm().invoke(enter=immediate).content material

Utilizing the operate is easy – simply move within the dataset, its title, and supply. The report format defaults to Markdown:

md_report = get_report_summary(
    dataset=df, 
    dataset_name="MBA Admissions (2025)",
    dataset_source="https://www.kaggle.com/datasets/taweilo/mba-admission-dataset"
)
print(md_report)

The HTML report is simply as detailed, however might use some styling. Perhaps you may ask the LLM to deal with that as effectively!

What You May Enhance

I might have simply turned this right into a 30-minute learn by optimizing each element of the pipeline, however I stored it easy for demonstration functions. You don’t must (and shouldn’t) cease right here although.

Listed here are the issues you’ll be able to enhance to make this pipeline much more highly effective:

• Write a operate that saves the report (Markdown or HTML) on to disk. This manner you’ll be able to automate all the course of and generate studies on a schedule with out guide intervention.
• Within the immediate, ask the LLM so as to add CSS styling to the HTML report to make it look extra presentable. You might even present your organization’s model colours and fonts to keep up consistency throughout all of your information studies.
• Broaden the immediate to observe extra particular directions. You may want studies that target particular enterprise metrics, observe a selected template, or embrace suggestions primarily based on the findings.
• Broaden the get_llm() operate so it could join each to Ollama and different distributors like OpenAI, Anthropic, or Google. This offers you flexibility to modify between native and cloud-based fashions relying in your wants.
• Do actually something within the get_summary_context_message() operate because it serves as the inspiration for all context information offered to the LLM. That is the place you will get artistic with function engineering, statistical evaluation, and information insights that matter to your particular use case.

I hope this minimal instance has set you heading in the right direction to automate your individual information reporting workflows.