Gemini sent me

[Barry Buckle]

So, I wrote a tool. it was actually for the automated mailing market but I got diverted (as you do) in to thinking about test data.

I've seen a lot of synthetic data tools recently that require "training" on real data. But what if you don't have real data yet? Or what if you need to test how your parser handles invalid data?

I built Synthed to be a lightweight, deterministic alternative. It doesn't use AI—it uses industry schemas. You get high-fidelity HL7, X12, and FIX data (and the ability to intentionally corrupt it) with zero privacy risk.

Check out the MIT repo: https://github.com/bbuckle1959/synthed

Test it, alter it, copy it - go mad! It's released under MIT licence

As for the post title, I asked Gemini where would be the best place to post this - and it suggested here.