Back to Guides

Cleaning Messy Data: I Turned 200 Junk Files into Gold in 10 Mins

2026-01-23 Akhil from Real Examples
Cleaning Messy Data: I Turned 200 Junk Files into Gold in 10 Mins

I opened my "scraped_data" folder yesterday and almost deleted the whole thing.

It was a graveyard of 200 text files I’d collected over the last year - notes, copy-pasted articles, half-finished ideas. I wanted to turn them into the "Blueprints" you see on this site.

> TL;DR: I had 200 messy files that needed to be cleaned up. Instead of writing a complex code script, I just wrote a simple "Memo" in English telling the AI what I wanted. It read the memo and fixed all 200 files in minutes. English is the new coding language.

But the data was garbage.

Some files used "Goal:" headings. Others used "Objective:". Some were just random bullet points.

To fix this the "right" way, a developer would write a Python script using Regex.

If you’ve never seen Regex, it looks like this: ^([a-z0-9_\.-]+)@([\da-z\.-]+)\.([a-z\.]{2,6})$.

Yeah. No thanks.

I am not learning that just to clean up some text files.

Prompt Engineering > Coding

Instead of writing code, I tried something dumb. I treated the AI like an intern.

I created a new file called MIGRATION_GUIDE.md and just... wrote in English.

I literally put:

> "Hey, look at this messy file. I want you to find the title (it's usually the first line) and the goal. Then rewrite the whole thing so it looks like this clean example below."

Then I ran the command:

> *"Read the guide. Then go through every file in the folder and fix them."*

It Actually Worked

I watched the terminal as it chewed through the files.

It was kinda spooky.

  • It found typos I missed ("Obejctive") and fixed them.
  • It guessed the category of the blueprint based on the text (tagging sales stuff as "Sales" and code stuff as "Engineering").
  • It formatted everything into perfect Markdown.

    • A rigid script would have choked on half these files. The AI just "got it."

    Why I'm Writing This

    I talk to a lot of founders who are sitting on piles of messy data - customer feedback, old leads, meeting notes. They think they need to hire a data scientist to clean it up.

    You don't. You just need to write a good set of instructions.

    We're at a weird point in tech where writing a polite memo to a computer is more effective than writing actual code. I'm here for it.