Jan Vincent
Szlang.

01 Research

We build, test, and select analytical engines based on assumptions about what they will actually run. Workload Intelligence replaces those assumptions with evidence.

I characterise analytical workloads systematically: at the level of the data they touch, the queries that run over it, and the deployments they live in. Grouping those characterisations into recurring types gives the field a shared vocabulary. For researchers, it clarifies which problems are worth working on. For industry, it turns production data into implementation decisions. For practitioners, it grounds OLAP engine choice in evidence.

• Thread
  01

  Do our benchmarks still reflect production?

  The benchmarks the field leans on were designed before today's production workloads existed. When the benchmark drifts away from what production systems actually do, research optimises for the wrong thing. I work on closing that alignment gap, and on the measurement tools that tell us whether it holds.

• Thread
  02

  Workloads have fingerprints. Engines should match them.

  Different analytical use cases leave fingerprints in the shape of their data and the queries that run over it. Different fingerprints justify different hardware, algorithms, and engines. I study how to characterise different use cases, and how to map their workloads to design decisions.

• Thread
  03

  A map of the analytical workload space.

  Beyond individual case studies, the longer-term goal is a taxonomy: the axes along which analytical workloads differ, the classes that recur in practice, and the optimisation levers each class unlocks.

02 Publications

• PVLDB 2025

  Workload Insights From the Snowflake Data Cloud: What Do Production Analytic Queries Really Look Like?

  Jan Vincent Szlang, Sebastian Breß, Sebastian Cattes, Jonathan Dees, Florian Funke, Max Heimel, Michel Oleynik, Ismail Oukid, Tobias Maltenberger.

  PDF ↗

• PVLDB 2025

  Unraveling the Impact of Window Semantics: Optimizing Join Order for Efficient Stream Processing

  Ariane Ziehn, Jan Vincent Szlang, Steffen Zeuch, Volker Markl.

  PDF ↗

03 Industry

• Snowflake
  Berlin

  Query Processing Engineer

  Built a vectorised expression framework. Co-authored the PVLDB 2025 paper analysing ~600M production BI queries. Profiled performance-critical code down to assembly.

• Huawei UK
  Edinburgh

  Senior Engineer, GaussDB

  Vectorised execution, in-memory data representation, selection vectors. Drove a 20% improvement on TPC-H through low-level optimisations.

• TU Berlin
  DIMA

  Research Assistant, NebulaStream

  Designed and implemented the cost-based window-join reordering algorithm published as the PVLDB 2025 window-semantics paper.