Back to Back

I would like to give a talk about this PostgreSQL extension: how we can use it in an AI pipeline with Python, and how to benefit from its «exactly-once» delivery guarantee

The Postgres write-ahead log, or WAL, is basically a changelog for the database. It enables several important Postgres features: crash recovery, point-in-time recovery, and binary and logical replication. This talk explains what is stored in the WAL, how binary and logical replication work, and how replication slots track replication progress.

Modern distributed systems need identifiers that are unique, scalable, efficient, and often sortable. In this talk, Mons Anderson, Solution Architect at Exness, will trace the evolution of identifier generation — from database autoincrement and UUIDs to Snowflake, KSUID, NanoID, ULID, and other modern schemes.

The talk will compare these approaches by uniqueness guarantees, sortability, storage efficiency, readability, operational complexity, and suitability for distributed systems. Special focus will be on UUID v7, which builds on lessons from previous generations and is becoming a new industry standard.

Backend development sometimes feels like detective work. One out of tens, thousands, or even millions of users encounters an issue. They might report it—or they might not. Either way, the developer has to figure out what happened, reproduce the problem, and identify its root cause.

Making that possible requires building observability into the system from the start. In this talk, I'll share several techniques we've implemented in systems operating at different scales to make user-reported issues easier to reproduce and diagnose. I'll also show how giving users better ways to share diagnostic information can significantly reduce the time needed to find and fix problems.

In real-time fraud detection, a failure doesn't always mean the service is down. The system may continue returning ALLOW while some of its checks are unavailable and confidence in the decision has decreased.

In this talk, we'll explore how to distinguish between VERIFIED, DEGRADED, and BLIND decision states, design risk-aware fallback strategies, preserve decision context for downstream consumers, and build effective post-decision verification workflows.