Original LinkedIn post

This week was all about JCrete®

It was hot, but fortunately that didn’t lead to any heated discussions. ;-)

I love this gathering. The people are not only extremely tech-savvy, but also simply kind.

It probably came as no surprise that AI was one of the recurring themes at jCrete this year.

Because of my day-to-day work, my academic time (yes, I studied AI at university years ago, before it became "fancy"), and the experiments I run in the evenings (sometimes far too late) I wanted to learn how others approach feeding data into AI systems.

So I proposed a session called "Data for AI". Why?

Here’s a fun fact: "garbage in, garbage out" applies not only to scientific experiments, but also to AI. Feed it with trash and expect gems later? ;-)

It is also funny how throwing all kinds of garbage into a "data lake" can eventually turn it into a "data swamp".

A few conclusions from the session:

• Bringing some order and structure to your data helps.
• Choosing the right models matters, both for different kinds of search and for reasoning.
• MCP might not be the best option for retrieval, especially when it is treated like CRUD.
• Benchmarking is your friend.
• Keeping an open mind doesn’t hurt either.

I’d like to thank everyone who made it possible for me to be there again this year: Heinz Kabutz, Ixchel Ruiz, Dmytro Vyazelenko (and the rest of the dis-organisers, especially the non-Kirks), and Elastic too.