How I Built a Masking Tool Without Showing AI Any Real Data: Column-wise Shuffling as the Scaffold

Short summary

A developer built a production data-masking tool by pairing AI assistance with strict data protection discipline: column-wise shuffling (Japan's official PPC method) plus Faker generation. The process never shared real data with LLMs—instead working in four phases of verification and refinement. Key discipline: survey industry terminology before AI implementation, which compressed code volume by 10x.

• •Built data-masking tool using column-wise shuffling (Japan PPC method) without exposing real data to LLMs
• •Followed four-phase AI collaboration: batch design, Excel verification, sample generation, Faker integration
• •Core discipline: survey industry standards before implementation, reducing code volume by 10x and ensuring compliance

Generated with AI, which can make mistakes.