How I Built MimAI, an AI File Assistant, With Claude Code

By Rapar Rubel (Rubel Miah) · 10 October 2026 · 2 min read

• MimAI
• Claude Code
• AI product
• building with AI
• Google Drive
• Bangladesh

Everyone has had this moment: you know the file is somewhere in your Google Drive, but you cannot remember the name, the folder or the date. You end up scrolling for ten minutes.

That small, daily frustration became MimAI — a file hub with an AI assistant built in. I designed it and built it with Claude Code. Here is what I learned.

What problem does MimAI solve?

People do not remember file names. They remember what the file was about and roughly when they used it. So instead of searching by name, MimAI lets you ask:

• "last month er invoice ta dao"
• "the property brochure I sent last week"
• "আমার CV এর নতুন version"

The assistant searches your own Google Drive and finds the right file — in Bangla, Banglish or English. (Why three languages matter is the subject of Why I Build AI That Speaks Banglish.)

Why did I build it with Claude Code?

I am a builder who thinks in products and businesses first. Claude Code let me work like a product lead with a tireless engineering partner: I described what the user needed, and together we wrote, tested and fixed the code.

It did not replace thinking. It made the distance between an idea and a working version much shorter, so I could spend my time on the decisions that matter — what the assistant should understand, what it should never do, and how it should feel to use.

What were the biggest lessons?

1. Start with one painful job. MimAI does not try to be a whole office suite. It does one job — find my file — and tries to do it well.

2. Describe the user, not the feature. "A shop owner who types in Banglish on a phone and is in a hurry" leads to better decisions than "add a search box".

3. Ship early, then improve. A live version that real people can try teaches you more in a day than a week of planning.

4. Keep secrets out of the code. Keys and passwords live in the hosting platform's environment settings, never in the repository.

5. Test with real requests. The best test cases are the strange, messy ways people actually ask for things.

How is MimAI deployed?

The code lives on GitHub and the app runs on Render, which redeploys automatically when I publish an update. It is a simple setup that lets a small team — or one person — ship improvements quickly and safely.

What does this mean for other businesses?

You no longer need a large engineering team to test a useful AI idea. If you have a clear, painful job and real examples of how people ask for help, you can build a working first version quickly — and learn from real users.

That is how I approach every AI workflow: small, useful, live, then better.

Want something like MimAI for your team? Tell me about it.