On this page
(Photo Credit: Berlin Buzzwords team)
Berlin Buzzwords is one of Europe’s leading conferences focused on search, data infrastructure, and machine learning—bringing together open source communities and industry practitioners to discuss what’s next in scalable systems.
What We Heard on the Ground
We spent more time listening to attendees, aiming to understand their current search setups and the challenges they faced. In our interactions, it was evident that most organizations attending were using Lucene-based distributions, predominantly Solr, followed by Elasticsearch and OpenSearch. These platforms have significant community know-how, and many attendees were already experimenting with vector search and hybrid search capabilities, though the experiences vary widely.
Engineers frequently expressed concerns around integrating dense vector retrieval into their existing Lucene-based stacks. Challenges included significant increases in index sizes and RAM usage, often driven by vector embeddings and immutable segment structures. Engineers also highlighted frustrations around partial updates, notably how updating a simple numeric field like price triggers full-document rewrites despite unchanged vectors, intensifying operational overhead. Many were grappling with performance tuning complexities, especially given latency introduced by additional reranking steps layered onto traditional retrieval methods. Lastly, navigating the nuanced ecosystem of freely available versus premium-only capabilities—particularly around features like reciprocal rank fusion (RRF)—added confusion to their decisions on architecture evolution.
Vespa's Fit and What Sparked Interest
Vespa has maintained a visible presence at Berlin Buzzwords since 2021, building familiarity among attendees over the years through various talks and production use cases—last year featured Mercari, and this year Climate Radar demonstrated its capabilities.
In our interactions this year, we noticed that while attendees were familiar with Vespa, many had not yet explored its detailed capabilities. The dedicated Vespa booth provided attendees an opportunity to dive deeper, showcasing practical demos such as ColPali for multimodal search and capabilities around e-commerce personalization. During conversations with visitors, we found that Vespa’s native multi-stage reranking, flexible experimentation through ranking profiles, and powerful tensor framework particularly sparked interest with the attendees. These features directly addressed some of the core pain points discussed—such as handling increased index sizes efficiently, managing reranking-induced latency, and simplifying complex experimentation processes—often leaving visitors pleasantly surprised by the possibilities within Vespa.
The Bigger Shift and Where It's Headed
We’re seeing a broader industry shift toward integrated, performant platforms purpose-built for AI-native search workloads. This trend only continues to grow stronger, and Vespa appears well-positioned to support these evolving needs.
At Searchplex, we're already seeing significant interest in multimodal, hybrid search, and personalization use cases—especially within e-commerce and Retrieval-Augmented Generation (RAG) scenarios. More teams are exploring migration to Vespa to gain flexibility and unify relevance, ranking, and retrieval. In that sense, we're here to help companies make that transition to truly AI-powered search systems with confidence and clarity. Learn more about our services here.
Marking a Milestone
Lastly, Berlin Buzzwords represented a meaningful milestone for us at Searchplex. While we've quietly focused on helping teams build fast, scalable, and intelligent AI-powered search systems, this was the first time we showed up publicly, sporting a tee and with flyers, to share what we do and engage with the community directly.
From quiet builders to visible presence.
Pictured here with the Vespa.ai team at their booth (L–R): Bonnie Chase, Radu Gheorghe, Piotr Kobziakowski, and myself.
Photo credit: Berlin Buzzwords team.