You get:
- manual data cleaning (slow, error-prone)
- inconsistent column naming
- missing value handling that’s not documented
- visualizations that don’t tell the right story
- analysis that can’t be reproduced
But data analysis is not one-off scripting.
It is a reproducible pipeline.
- Load data: from CSV, Excel, SQL, or API
- Clean data: handle missing values, rename columns, filter outliers
- Transform data: group, aggregate, pivot, merge
- Analyze: statistics, trends, correlations
- Visualize: charts that answer specific questions
Without a structured approach, analysis is messy.
This framework forces AI to generate reproducible analysis scripts.
Assume the role of a data analyst who writes clean, reproducible pandas code. Your task is to generate a data analysis script. Generate: 1. IMPORTS - pandas, numpy, matplotlib/seaborn 2. DATA LOADING - Read data from source - Initial inspection (head, info, describe) 3. DATA CLEANING - Handle missing values (drop or fill) - Rename columns (snake_case) - Fix data types - Remove duplicates 4. DATA TRANSFORMATION - Filtering, grouping, aggregating - Creating new calculated columns 5. ANALYSIS & VISUALIZATION - Summary statistics - Key charts (histograms, bar plots, scatter plots) 6. INSIGHTS SUMMARY - Key findings in plain English INPUTS: Data Source Description: [E.G., "CSV file with columns: date, product, sales, region"] Data Cleaning Needs: [E.G., "Dates are strings, sales has missing values, region has typos"] Analysis Questions (2-3): [LIST] Desired Visualizations: [E.G., "Sales by month," "Sales by region," "Top 5 products"] Output Format: [JUPYTER NOTEBOOK / PYTHON SCRIPT] RULES: - Use pandas for data manipulation - Use matplotlib/seaborn for visualization - Add comments explaining each step - Handle common data issues (missing values, wrong types) - Make visualizations publication-ready (labels, titles, legends) - Print summary statistics before and after cleaning
- Describe your data schema clearly — column names, types, meanings.
- List specific cleaning needs — missing values, outliers, typos.
- The analysis questions drive what visualizations to create.
- Test the generated script on a sample of your data first.
- Add your own domain-specific logic after generation.
Data Source Description: CSV file with columns: Date (string), Product (string), Sales (number), Region (string: East, West, North, South)
Data Cleaning Needs: Date column needs conversion to datetime, Sales has missing values (fill with median), Region has typos (“Eat” instead of “East”)
Analysis Questions: “Which month had the highest sales?”, “Which region has the best performance?”, “What are the top 3 products?”
Desired Visualizations: Sales by month (line chart), Sales by region (bar chart), Top 5 products (horizontal bar chart)
Output Format: JUPYTER NOTEBOOK
This framework improves outcomes by forcing:
- reproducible pipeline (load → clean → transform → analyze)
- explicit cleaning steps (handles edge cases)
- visualization best practices (labels, titles, legends)
- insights summary (plain English findings)
- pandas best practices (vectorized operations)
Great data analysis scripts don’t just produce numbers — they tell a story and can be rerun.
Build Better AI Systems
Subscribe for advanced prompt engineering, AI coding tools, Python frameworks, and practical strategies for developers and engineers.