SEON TechTalks
Excited to announce the very first edition of SEON TechTalks, where we’ll uncover cutting edge engineering topics through hands-on sessions!
At the first event, we’ll dive into two of the hardest engineering challenges modern product teams face at scale: building flexible systems without losing performance, and navigating the hidden complexity of distributed architectures.
Viktor Micskó (Staff Software Engineer at SEON)
When Your Users Write Your Queries: Operating a Fraud Detection Rule Engine on PostgreSQL at Scale
SEON's fraud detection platform lets customers define their own scoring rules, from simple threshold checks to rolling-window aggregations across millions of transactions, all evaluated in real time. This flexibility is the product's strength, but it creates an unusual engineering challenge: customers are effectively authoring the database queries, and with over a million rules across thousands of customers and multiple PostgreSQL shards, performance becomes fundamentally unpredictable.
This talk traces how we evolved from reactive firefighting to a systematic, increasingly automated performance engineering pipeline. We'll start with the constraints that make this problem hard (a customer-configurable rule engine on sharded PostgreSQL with real-time latency requirements), then walk through how a large-scale rule analysis shaped our understanding of real-world workloads.
From there, we'll show the observability layer we built on ClickHouse, the toolchain it powers (automated index recommendations with hypothetical testing, merged query plan visualization, performance dashboards), and the weekly performance review process that ties it all together. We'll close with how we're now automating that entire workflow with AI agents, moving from a human-driven weekly ritual to a continuous, autonomous performance loop.
István Ignácz (Staff Software Architect at SEON),
Accidental Distributed Systems - The Art of Impossible
You've probably never implemented Raft, designed a consensus protocol, or worried about Byzantine fault tolerance. But if your application talks to a database, publishes messages to a queue, or calls an external API - congratulations, you're running a distributed system. You just didn't mean to.
In this talk, we'll look at the "impossible" problems hiding in everyday software: saving to a database and emitting a Kafka event in the same request, deploying a new version without a moment where old and new code coexist, or processing a message exactly once. These aren't edge cases - they're fundamental impossibilities rooted in decades of distributed systems research. And yet, we ship software every day - maybe not Friday.
We'll explore why these problems are truly impossible, why that matters for the code you write on Monday morning, and what pragmatic techniques let you dance around the impossible and build systems that work well enough.
Schedule
18:00-18:30 - Arrival
18:30-19:00 - Viktor Micskó (Staff Software Engineer at SEON) — When Your Users Write Your Queries: Operating a Fraud Detection Rule Engine on PostgreSQL at Scale
19:00-19:30 - István Ignácz (Staff Software Architect at SEON) — Accidental Distributed Systems - The Art of Impossible
19:30-21:00 - Drinks, Pizza, Networking 🍕