datawhisperer 0.1.4

Natural language interface for DataFrames using LLMs

🧠 DataWhisperer

Talk to your DataFrame. Literally.

DataWhisperer is a professional-grade Python library that enables interaction with pandas DataFrames using natural language. Powered by LLMs like OpenAI and Google Gemini, it transforms analytical questions into executable Python code. Whether you need summaries, transformations, or visualizations, DataWhisperer delivers accurate results and adapts dynamically to your data structure.

✨ No more .groupby(), .pivot() or plotting boilerplate. Just whisper to your data.

---

🚀 Key Features

• 🔗 Natural Language to Python code conversion for DataFrames
• 📊 Auto-generated visualizations with Plotly
• 🛠️ Automatic error detection & self-repair with multi-step retry logic
• 🧠 Supports OpenAI, Gemini, and is extensible to Claude, LLaMA, Mistral (coming soon)
• 🔁 Smart retry strategy for broken code (max_retries)
• 🧼 Schema-driven prompt system (auto-describes your columns)
• 🧪 Modular and testable architecture
• 🧩 Plugin-ready and LLM-client agnostic (future-proof)

---

📦 Installation

Install from TestPyPI:

pip install -i https://test.pypi.org/simple/ datawhisperer

Or, for local development:

git clone https://github.com/JosueARz/DataWhisperer.git
cd DataWhisperer
pip install -e .

Requirements: Python 3.8+

---

⚡ Quick Start

from datawhisperer import DataFrameChatbot
import pandas as pd

# Load your dataset
data = pd.read_csv("sales_data.csv")

# Describe the columns
schema = {
    "region": "Sales region (e.g., North, South, East, West)",
    "sales": "Amount of revenue generated",
    "date": "Date of the sale"
}

# Create the chatbot
bot = DataFrameChatbot(
    api_key="your-api-key",
    model="gpt-4",  # or "gemini-1.5-flash"
    dataframe=data,
    schema=schema,
    max_retries=3  # Number of auto-repair attempts if execution fails
)

# Ask a question
response = bot.ask_and_run("Show a bar chart of total sales per region")

# Display result
print(response.text)
response.table  # Or: response.chart

---

🧠 What kind of questions can I ask?

• "Which region had the highest revenue in Q2?"
• "Show average sales by month."
• "Plot a heatmap of transactions by region and month."
• "How many sales were made after July 15th?"
• "Which regions had declining revenue trends over time?"

DataWhisperer will generate optimized and runnable Python code — and automatically fix it if it fails.

---

✅ Example Output

🔍 Input (user question):

"What are the top 3 regions by total sales?"

🧪 Output (Python code):

import pandas as pd
import plotly.express as px

# Group and sort data
top_regions = df.groupby("region")["sales"].sum().sort_values(ascending=False).head(3).reset_index()
print("Top 3 regions by total sales:")

# Display table
top_regions

📋 Output (text + table):

Top 3 regions by total sales:

region	sales
North	102,400.0
East	98,900.0
South	87,120.0

---

🧪 Testing the Library

pytest
pytest --cov=datawhisperer --cov-report=term-missing

Coverage Includes:

• Code generation & formatting (PromptFactory)
• Code execution & error repair (executor.py)
• Retry logic & self-healing (fixer.py)
• Chat interface orchestration (DataFrameChatbot)

Note: Tests use fake LLM clients to avoid real API calls.

---

📌 Version

Current release: v0.1.4

See the CHANGELOG.md for updates.

Upcoming versions will include:

• SQL + Python-SQL modes
• Multiple LLM client support
• DuckDB, SQLite, Postgres connectors
• CLI + Streamlit interface
• Plugin hooks

---

📖 Documentation

Full documentation will be hosted via MkDocs in future release v1.0.0.

• Quickstart & API Reference
• Examples & Recipes
• Plugin Development Guide

---

📄 License

Apache License 2.0 — use it freely, contribute professionally. © 2024 JosueARz

---

🤝 Contributing

Pull requests are welcome. Please ensure your changes are:

• Type hinted and PEP8-compliant
• Tested with pytest
• Explained clearly in PR descriptions

---

🧙 Why "Whisperer"?

Because it doesn’t just chat — it understands your data.