빈 데이터베이스 문제: 하나의 명령에서 현실적인 테스트 데이터

You finished the schema. The models are in, the first screen renders. Then you go to build the real thing and there is nothing there. The table is empty, and you cannot test a feature against zero rows.

So you improvise. A few hand-written fixtures. A prompt to a chatbot that returns five rows, two of them near duplicates, one quietly breaking your schema, and none of them the same when you ask again tomorrow. You came to build a feature, and instead you are maintaining fake data.

Here is the part most people miss: the expensive thing about test data is not the rows. It is deciding what a good row looks like. Do that thinking once, capture it, and you never pay for it again. That is the whole trick.

That is what a template is. You describe the data you want in a sentence, and you get a generator you can run forever:

pip install dugalaxy
dugalaxy gen customer-support

Enter fullscreen mode Exit fullscreen mode

customer-support ships in the box. Describe your own in a sentence and you get your own template. Either way the output is varied, consistent, and reproducible: same seed, same data, every time. Tomorrow’s run matches today’s, which means your tests stop lying to you.

Three things that make it hold up:

It is an asset, not a batch. The template captures the judgment once and repays it on every run.

It is grounded, not improvised. Samples are generated against deterministic structure, so the fields line up and you can validate the shape. A chatbot re-rolls the dice on every call. A template does not.

It runs on your machine. No account, no upload, offline. Your data stays yours.

And the honest limit, because you should hear it from me first: it validates structure, not meaning. It checks that a sample fits the shape and rules you defined. It does not read prose and swear it is semantically perfect. A generated template is a strong starting point you still tune, not an oracle. The boring, repetitive part is handled, and your judgment goes where judgment actually matters.

A chatbot gives you five samples that drift. A template gives you data that survives.

If you have ever stared at an empty table thinking “I just need realistic data so I can build this,” that is the entire reason this exists.

pip install dugalaxy. Fill the empty database, then tell me what breaks.

Repo: https://github.com/nugalaxy/dugalaxy · Built at https://nugalaxy.ai