Schedule

A year ago, there was the first survey and the first results of the State of Data. This time we will not just look at the results, but also see the dynamics: what has changed over the year.

We will discuss the key challenges that Apache Iceberg is facing, as well as the prospects for technology development.

I will reveal how codecs LZ4, ZSTD, Delta, and DoubleDelta help increase query speed and reduce storage volume. I will highlight the challenges that arise when using them in projects.

How we at T-Bank built our BI tool on Apache Superset, rebuilt our BI culture, made synergies between BI analysts and developers of our BI tool and successfully migrated from Tableau.

I will tell you about a method for automated optimization of Apache Spark configuration for ETL processes using Spark metrics and the RAG system, which significantly optimizes the utilization of ETL processes.

I'll tell you how we use Airflow in practice: from the pain of sensors to the convenience of datasets, from standard features to our own custom solutions. The talk will not leave those who are faced with the actual operation of Airflow indifferent.

A practical case study of implementing DWH monitoring from Skyeng: from metadata architecture to automated data quality checks and transition to DataOps practices.

Let's talk about Spark. What did it give data engineers? Why do many of us use it?

Spark has been around for over 15 years. What problems do we face when using it? Is there anything better? Is it already possible to replace it with something?

Why is %SQLEngineName% slowing down? How can one fix this? Benchmarks, open source, and the like.

Description of the path of developing an open source data lineage solution based on OpenLineage. Comparison with other open source solutions — OpenMetadata, DataHub, Marquez — and the reason we abandoned them in favor of our own development. No, this is not another custom Data Catalog :)

How to build a secure, powerful, and scalable LLM service for a large company: with UI, API, moderation, and model support for completely different tasks.

We upload hundreds of terabytes from Greenplum to S3 every day. You can learn about the pitfalls we have collected and what happened in the end.

How Yandex Market started writing documentation. You will learn how it happened and what problems the company faced. We will consider different approaches to describing metadata in storages, compare them with each other and understand whether it is worth going down this path.

A detailed review of existing vector search algorithms, the most popular in modern database management systems.

The story of how a small team of engineers implemented Hadoop with full Kerberos and Ranger-based security without stopping business processes.

I'll tell you about an AI assistant that helps users get answers to questions about data. You'll learn how we at X5 Tech manage the quality of answers and how data and data descriptions affect the final result.

YDB has undergone a significant development path from applying basic vector search techniques to creating a scalable and efficient vector index. The talk presents a detailed analysis of the stages of evolution of vector search in YDB, including analysis of complexities and engineering solutions. 

In this talk, I will tell you how we migrated from a platform based on Vertica, HDFS to the new Dota 2 (the second version of our internal analytics platform)) architecture based on Apache Ozone (S3), Trino, Spark and Iceberg. I will share our experience in choosing storage, explain why we abandoned HDFS and why we chose Apache Ozone as an on-prem implementation of S3.

Good data doesn’t happen by accident. I’ll share my experience building a tool that helps validate data automatically — fast, flexible, and pain-free.

How Data Lakehouse became our lifeline: painless migration with a continuous flow of more than 150 TB per day.

I will tell you about Spark Connect — a new approach to working with Apache Spark, which allows you to develop the client part of the application in any language and not depend on the JVM. We will talk about the architecture of Spark Connect and its differences from classic Spark. You will learn about a project where we used Spark Connect API for C++.

How to build meaningful Retrieval-Augmented Generation (RAG) pipelines where LLM doesn't just “guess” the answer based on similar chunks, but consciously explores the data based on its structure and relationships.

Review and comparison of existing Python libraries and a self-written profiling tool for data quality analysis. Description of the tool's functionality.

The architecture of the testing and deployment service for showcases in Avito and the approaches used in testing showcases.

Getting change events from sources is quite a common task that can be solved in different ways. One of such solutions is Debezium. But is it so simple and is it always the best solution? I will try to answer these questions and consider Debezium from the point of view of the difficulties that arise on the way of solving the task of change capture.

Approaches to uploading metadata to the Data Catalog are often considered in a linear way: a minimum of changes, maximum preservation of the "truth". But is this really the right thing to do?

The data platform in our company has existed for more than 5 years, during this time it has absorbed a lot of trendy (and not so trendy) solutions. I will tell you how we tried to choose our future among ClickHouse, Greenplum and Trino, and found StarRocks. 

We'll talk about how Wildberries implements a JupyterHub and Kubernetes-based research platform for more than 600 data scientists who solve problems in areas such as CV, NLP, OCR, and recommendations.

The potential of using AI to automate Data Governance processes on the side of data platform users.

How the T-Bank team migrated DataOps to Kubernetes and didn't go crazy. I'll tell you how we designed and implemented infrastructure for managing the lifecycle of ETL tasks using Kubernetes operators, automated DAG delivery and integrated it into the existing DataOps. I'll analyse what happened, where we made mistakes, and what you absolutely shouldn't do.

How Yandex managed to bring order to the chaos of distributed data using an internal data contract service — without centralization, but with clear responsibility and transparent agreements.

The database is already covered with read replica, but it is still not eniguh — what should you do?

I'll tell you in detail about our experience with YugabyteDB, which we chose as the solution. We will discuss important settings, nuances from the point of view of development and bugs that we found.

For those who will be rolling YugabyteDB into production, the talk will save a lot of time and nerves. But it will also be interesting for those who use PostgreSQL or another classic relational database and are thinking about its scalability and fault tolerance.

Experience of using Comet and Gluten (Velox) execution engines – from the introduction and features of the build to the results of testing on real ETLs. I will tell you about pitfalls and non-obvious points, show the results of work and consider cases when these engines are useful and when they don't work at all.

The real experience of building DWH in StarRocks: architecture, application cases, pitfalls. Whether StarRocks met our expectations or not.

How to implement a Data Quality distributed architecture tool that ensures smooth operation for a large number of teams and is a single point of truth about data quality in company systems.

How to configure and modify Apache Spark for your tasks without rewriting the framework. I will tell you about approaches to expanding the functionality of Spark SQL without interfering with the platform's source code. You will learn about creating your own data sources, developing user functions for specialized processing, and implementing optimization rules that adapt to various requests.

How can we measure the quality of a data platform and manage its development? I'll tell you how at Yandex Delivery we built a metrics system to evaluate 7 key areas — from infrastructure stability to business data usage.

What metastore is, how it works in the big data ecosystem, what solutions exist on the market and why we decided to develop our own. I will share practical experience, architecture and lessons we have learned.

How the connection between nature and man helps to solve a variety of tasks.

We will be summarising the results of the conference, recalling the highlights and talking about future plans. Join us in the hall or online so you don't miss a thing!