Synthetic data generation for email summarization tool testing

Currently I am working with some students on a project (let's call it tool-1), it is being designed to have the following features

1. Pick a set of emails based on sender, timestamp, etc.
2. Let user select important themes/projects for that week
3. Remove toxic content from emails
4. Summarize the emails based on themes/projects

If done right, it would be very attractive tool for senior executives.

The problem that we were facing was that how to test the solution?? Then a friend of mine gave an idea!? The idea was that we develop a tool (lets call it it tool-2) which takes summarized emails with the above qualities (especially the themes/projects part), then it would split them into multiple emails.??

We will then use tool-2 to generate lots of emails and test tool-1.? If we can arrive at the same/similar summary then we have a reasonable way to check the quality of tool-1.?

Shows the power of synthetic data, just like the AlphaGo team got their Machine Learning algorithm to play against a different version of the same algorithm.? This way they were able to play these algorithms against each other millions of times, coming up with moves that the grand master also did not know about.

Do reach out if you want to discuss more ideas around synthetic data generation.