A week has gone by now since Haystack 2024 EU took place in Berlin. Unfortunately, I was not able to attend in person, but as the organizers made it available to join the sessions remotely, I happily marked the days in the calendar, eager to listen about the latest innovations in information retrieval and case studies from fellow search engineers.
Haystack is an annual conference with installments in Europe and the US organized by OpenSource Connections. The focus of the conference is on information retrieval and search technologies. Of course you can hear about the latest trends in the field, but that is not what the conference is all about: There is plenty of room for actual practitioners to tell their story, and - speaking for myself - this is where I get the most value out of the conference: Hearing about their journey, probably facing similar problems than myself or learning about what obstacles to expect when trying out the new stuff and how to overcome them. And, with that being said, this year's EU installment of Haystack EU did not disappoint at all.
The conference was started off by Trey Grainger and his keynote on AI-Powered Search. As generative AI makes its way into search technologies at a rapid pace, it is not easy to keep up with the latest developments. Trey managed to provide a great overiew of the latest trends and innovations in the field, especially with a focus on AI-driven features. So, if you do not know what vector databases are all about, what the purpose of a bi-encoder is or if you never heard of stuff like Colbert or ColPali, I encourage you to watch that session once the videos are available to the public. By the way, Trey is also writing a book (same title) on the subject.
AI was a big part of the conference in general, but not its sole focus. While especially LLMs can help you to build MVPs faster and lead to quicker improvement cycles, there was also the notion that we should not forget about all the non-LLM changes that are battle-tested and enable us to improve the quality of experience for search as well. In their talk Search and LLM - Building a GenZ Search Application the folks over at Knowunity demonstrated this quite nicely by showing how they experiment and evaluate in order to improve the quality of their product (granted though, their talk was leaning more towards the AI-perspective of that discussion). For the more classical approach, the talk from Julien Meynet from Wallapop on Evaluating E-commerce and Marketplace Search did stand out in my opinion. Julien talked about why we need to collect user and business metrics, how to do it and why it is key to understanding their significance to build (better) search systems. There are a lot of great insights and takeaways from that session, highly recommended to check it out.
Especially when it comes to the collection of user metrics, search technologies like OpenSearch can decrease the barrier of implementing a user-driven feedback channel. Aswath Srinivasan from Amazon Web Services talked how to leverage the capabilities of OpenSearch to capture what users are doing in his talk User Behavior Insights to Enhance Search Relevance. Even if you are not using OpenSearch yourself, but are keen on learning how others are acquiring that data, I believe that this talk can be considered as a good starting point to get some ideas.
There is a lot more of interesting stuff that I will not unpack here. If you did not get the chance to attend the conferece yourself but are interested in the topic, there is good news though: The Haystack organizers will open up recordings of all sessions to the public a couple weeks after the conference. I highly encourage you to check them out.
Shoutout to all the good folks over at OpenSource Connections for organizing the conference. It has been a great experience!
