J-Fall 2023: What if AI generated your test data? - Wim Blommaert
Synthetic (fake) images made by generative AI have been used in the medical world to train
tumor detection software for several years now.
Recently it became possible to generate synthetic tabular data, which makes up the bulk
of the data in a bank like ING.
In software Engineering, testing is a very important activity but its only as good as the available
test data.
The automatic generation of synthetic tabular test data can help engineers so
they can focus on producing code.
At ING we started the journey to use Generative AI to create Synthetic tabular data more than
2 years ago. I’m running a dedicated team that has already connected 11 applications to synthetic
data and is now shifting focus towards the scaling of this technology cross the bank.
I will use a few examples of real projects to showcase what this is going to do for Software
Engineering.
In one such project we are now able to generate 10k synthetic payments (SEPA) in less than
1 minute which manually took days to just create a few.
In another one we generated thousands of synthetic request / reply json pairs which we then use to
mock REST API’s.
Or we produced a huge dataset for performance testing by an external party.
We are also looking into the integration of this technology into Azure deployment pipelines in
support of test automation.
And recently we have started to look into the use of synthetic data for data science projects.