This presentation is designed for engineers, architects, and technical leaders who want not only to use large language models, but also to understand how they work, how to interact with them through APIs, what challenges arise when building RAG systems, and how to solve them.

During the talk, we will explore why a simple RAG approach is no longer sufficient for organizations with multiple data sources and how modern AI systems are evolving — from classic RAG to Deep Search, hybrid search, and Knowledge Graphs as a layer of corporate memory. We will also look at how to combine unstructured documents, tabular data, databases, internal wikis, chats, and business entities into a unified system where AI can find relevant sources, build a search path, explain relationships between facts, and provide more accurate and verifiable answers.

With the rapid development of AI tools, the IT industry has reached a point where building a new project is often cheaper and easier than maintaining an existing one. However, for many large, long-running projects, rewriting everything from scratch is not the best option — especially if teams learn how to maintain and evolve them with the help of agentic tools. In this talk, we will explore real-world experience working with such projects, covering key aspects of collaboration with people, tools, technologies, and processes through practical case studies.

Most engineers eventually face the need to perform load testing: validating how a service scales, testing a new database, or running performance benchmarks for a new technology. At that point the obvious question arises — which tool should you use? Existing solutions work well for HTTP load testing, but they often become limiting when you need to test other protocols, model complex workload patterns (open vs closed systems, skewed distributions, hot partitions), or run distributed load testing in a cluster. In this talk, I will introduce NBomber — a load testing framework I created to address these challenges. We will cover: - why there was a need to build a new tool despite the existence of Gatling, Locust, and k6 - using .NET and F# to build latency-sensitive systems - the architecture of NBomber - how NBomber Cluster works - several practical use cases including database benchmarks, anomaly detection, Kubernetes integration, benchmark comparison, and performance trend analysis.

Let’s be honest: in most teams, performance optimization only becomes a priority when something is already on fire. And even then, it’s usually handled by one or two people. Not because others don’t care — but because real optimization requires a lot of time, context, and expertise. At Temabit, we decided to experiment with a different approach: delegating part of the optimization work to agents. In this talk, I’ll share what came out of it: how we learned to frame optimization tasks so agents can produce useful results, how we validate their suggestions, and why optimization turned out to be much harder to delegate than writing code — but potentially far more valuable. Real cases, honest lessons, no hype. And one key question: how realistic is it to trust agents with the performance of your product?

If engineers don’t think like product managers, PMs become the bottleneck. In many teams, engineers execute tasks while PMs carry strategy, context, and decisions alone. The result? Slow innovation, low ownership, and “feature factory” delivery. This talk explores we explore how PMs can redesign the system - not just the mindset - to build outcome-driven teams. You’ll learn: • Why feature factories emerge • How PM overload happens • How to push context and decisions down • How to create shared ownership of metrics and impact • Real examples of transforming delivery teams into product teams

AI development hits everyone, so hit us. Everybody wants AI agents to replace regular UIs. In this talk, I will describe the evolution of our multitool AI agent, built with Node.js on top of Google Vertex AI. I’ll dive into our journey of choosing the right models and scaling development through CI/CD, TDD, and performance monitoring. Is it even possible to achieve stable results for AI projects that can hallucinate and return various responses? Interestingly, we eventually decided to remove MCP servers and Zod schema validation—technologies often considered the "standard" for these tasks. Want to know why we moved away from them? Join my session to get these insights and ask your questions live!

- Professional deformation and its consequences. - Market bias and perception — how to find a balance? Let's talk about two important components of any experienced product leader's profile: their expertise, which can become a burden, and the difference in the dynamics of the development of that expertise and the market to which it relates. There will certainly be more questions than answers, but the search for answers will be an interesting adventure for your “homework.”

AI video generation services are evolving extremely fast, effectively opening up a new way of creating video — at a moment when the standards, language, and rules of this industry are still being formed. It feels like a reinvention of cinema: there is no single “right” approach yet, but it’s already clear what prevents generation from becoming a controlled production process. I will share ideas on how existing platforms can be improved, drawing on directing experience: which director-level needs a product should account for, where modern services most often break the pipeline, and what product changes could significantly increase the quality of the final result. The focus is on a platform that works equally well for two audiences: experienced filmmakers who need control and predictability, and “native AI” creators who started their journey in video through generation and have no traditional production background. For both groups, I will outline the logic of a pipeline: "idea → generation → rough edit → export", including the minimum viable editing inside the service and professional export to Premiere or DaVinci for finalization.

Talking about something abstract is easy and fun — you can sound smart, and no one can really verify your claims because, well, it’s abstract. In a world where every other person is now an “AI visionary,” I want to speak about AI in a максимально concrete way, using my own work at GitLab as the example. It’s a perfect case. While many people describe a world of pink ponies — greenfield projects built from scratch — I want to share the hardcore brownfield reality: what actually happens when you introduce AI into a project that has existed for years. Here’s what we’ll cover: - How AI and JavaScript get along (or, more accurately, don’t) - My setup and “secret sauce” for agentic development - What I’ve achieved together with Opus (my favorite model) - How much JavaScript code I still write by hand - What prevents us from reaching the mythical 10x productivity — and what I’m doing about it My goal is simple: after this case study, everyone should walk away with a clear understanding of what they can implement in their own project to make life better — starting now.