Meetup: Vector databases, RAG, BM25, sparse searches, AI model training

Vector databases, RAG, BM25, sparse searches, judging AI training output efficiency… Do you know what each of these topics means? That’s exactly what we covered at the latest “Bavaria, Advancements in SEarch Development” (BASED) Meetup.

The speakers shared a bunch of tricks and strategies from their ongoing projects, and I can confirm that it was packed with cool and practical stuff.

Clelia Astra Bertelli presented multiple use cases for vector databases and shared some of her own PoCs and project repositories. At the end of the event, she also answered a few questions I had about tools I’ve been reading up on. Super helpful.

Right after that, Evgeniya Sukhodolskay walked us through her research on semantic improvements for BM25. It was great to learn more about sparse vs. dense data approaches.

Then, Monica Riedler shared her publication on improving RAG by using multimodal inputs and strategies on how to evaluate each model’s outputs after adding a new modality.

Finally, Daniel Wrigley introduced Quepid a really handy tool for evaluating how good your search results or prompts actually are.

It was an amazing meetup lots of real-world techniques and tips, and the chance to meet people who really know their stuff.

A special thanks to Tacto for hosting the event, and to Daniel Wrigley and Evgeniya Sukhodolskay for organizing it. Everything was perfect. Looking forward to the next one!