Schedule

This is not a talk on the successful adaptation of AI, nor is it a set of recipes on how to integrate AI into your team. Instead, this is a revelation: I will honestly share my experience as a developer in the era of AI, the conclusions I have drawn, and my predictions for the future.

The hidden risks of implicit actions when building Android applications and the implementation of the shift left approach for early error detection through automated checks.

What has changed, how to work with it, and what are the prospects for using Swift at Android.

How to solve the problem of different development speed in a large mobile application.

We have implemented an AI assistant in the Drinkit mobile application as a new interface for a complex menu: the user says "I want something refreshing", and the application selects a drink and customization.

The talk includes an analysis of the code and architecture: Firebase AI Logic, function calling + agent loop, how not to go bankrupt, and an alternative via Koog (if Firebase is unavailable).

We dive into the "complexity of the debug" of SwiftUI using the example of incorrect operation of the GeometryReader and LazyVStack bundle. We analyze how AttributeGraph participates in calculating the size and updating frames, and form a practical approach to diagnosing and solving such problems.

A 700,000-line mono project—why is it cool and how to live with it?

In the talk, we will analyze how we implemented mobile pipeline failure point monitoring at Tutu, learned how to automatically classify the causes of crashes using AI, and divided responsibility into Dev, Core, and External levels. I will show you how the introduction of pipeline stability SLAs and nightly reports from LLM help not only resolve incidents faster, but also plan infrastructure development for the coming quarters.

The talk is about the work of isolates and the garbage collector in Dart VM, the connection of isolates with OS threads and restrictions on their number.

We will dive into the Codable internals, find bottlenecks, design a solution, and commit it to Swift itself!

Let's look at what smoothness is in a mobile app and what metrics to measure it with (and why FPS alone is often not enough). Let's see a case study: how measurements found a bottleneck in the product scenario and improved the metrics.

A session in the format of a level-by-level game will show how algorithmic interviews work and what skills need to be upgraded in order to pass each stage of the selection process.

Shared animations are increasingly becoming not just a decoration, but a required part of the design.

I'll share how we at Drinkit implemented unusual Shared Transitions with blur in Compose and why there's no reason to be afraid of this instrument.

This talk demonstrates why using Metal does not always lead to performance improvements. Using an image-processing example, we’ll explore when the GPU truly wins and how to measure it correctly.