Realtime Magic: The Go-Powered Engine Behind Millions of Decisions Per Second / Or Yarnitzky

Arpeely operates within the fast-paced AdTech ecosystem, handling millions of ad requests per second while maintaining sub-300ms latency. To meet the demands of high-performance, low-latency decision-making, we have developed an extremely configurable realtime engine that adapts dynamically to evolving business needs. In order to master such high scale and performance constraints, our main services are written in golang and leverage the expr-lang Go expression library (https://github.com/expr-lang/expr) to execute dynamic logic configured by data scientists in real time. This approach allows for rapid iteration without requiring deep engineering involvement, empowering data teams to optimize decision-making pipelines efficiently. We have implemented key optimizations to enhance the expression library’s efficiency, including expression caching, lazy variable resolving, and parallelism enabling us to process complex logic at peak efficiency. In this talk, we will dive into the technical challenges we faced, the innovative solutions we developed, and how dynamic expression engines in golang can reduce the need for strong coupling between engineering and data science teams.