Claude Code Data Analysis

How I Use Claude Code for Data Analysis (Without Exposing My Real Data) ☕

Hey there! 👋 This repo contains all the code, data, and prompts from my YouTube video on doing privacy-first data analysis with Claude Code.

🎥 Watch the full video tutorial here: https://www.youtube.com/watch?v=IPKOx66fogE

If you work with real-world datasets (like customer records, employee data, or financial transactions), you already know you can't just upload your CSV to an AI tool and hope for the best. Even running an AI coding agent locally in your terminal still sends data to the cloud to process your requests.

But here's the thing: as a data analyst, you don't actually need AI to analyze the data *for* you. You just need it to write solid, working analysis code that you can inspect, run and test yourself.

🛠️ How the Workflow Works

In this demo, we pretend we're analyzing 2,000 guest records from a boutique hotel (hotel_guest_data.csv), which includes sensitive PII like guest emails, billing addresses, and credit card numbers *(don't worry, the "real" dataset in this repo is 100% fictitious!)*.

Here is the 3-step workflow we use to keep the original data completely out of reach of the AI:

1. Create a "fake twin" locally (with SDV):
   Using a Python library called Synthetic Data Vault (SDV), we learn the statistical patterns of our local dataset and generate a 100-row synthetic proxy (hotel_guest_data_synthetic.csv). It has the exact same columns, data types, and structure, just with fake emails, addresses, and card numbers.
2. Let Claude Code do the heavy lifting in the cloud:
   We open Claude Code in Cloud mode (so it has zero access to our local files) and upload *only* the synthetic CSV. Claude writes and tests the exploratory data analysis notebook (pandas + Plotly) and…