About Nebius AI

Launched in November 2023, the Nebius AI platform provides high-end infrastructure and tools for training, fine-tuning and inference. Based in Europe with a global footprint we aspire to become the leading AI cloud for AI practitioners around the world.

Nebius is built around the talents of more than 500 highly skilled engineers with a proven track record in developing sophisticated cloud and ML solutions and designing cutting-edge hardware. This allows all the layers of the Nebius AI cloud – from hardware to UI – to be built in-house, clearly differentiating Nebius from the majority of specialized clouds: Nebius customers get a true hyperscaler-cloud experience tailored for AI practitioners.

As an NVIDIA preferred cloud service provider, Nebius offers the latest NVIDIA GPUs including H100, L40S, with H200 and Blackwell chips coming soon.

Nebius owns a data center in Finland, built from the ground up by the company’s R&D team. We are expanding our infrastructure and plan to add new colocation data centers in Europe and North America already this year, and to build several greenfield DCs in the near future.

Our Finnish data center is home to ISEG, the most powerful commercially available supercomputer in Europe and the 19th most powerful globally (Top 500 list, June 2024). It also epitomizes our commitment to sustainability, with energy efficiency levels significantly above the global average and an innovative system that recovers waste heat to warm 2,000 residential buildings in the nearby town of Mäntsälä.

Nebius is headquartered in Amsterdam, Netherlands, with R&D and commercial hubs across North America, Europe and Israel.

The role

We are looking for a Software Engineer to join our TractoAI team. TractoAI is built on top of the mature, open-source YTsaurus technology, which boasts a decade-long proven track record within BigTech companies. Our platform is designed to be highly scalable, resilient, and multitenant, capable of handling up to tens of thousands of servers, exabytes of data, millions of CPU cores, and tens of thousands of GPUs. TractoAI is a new experimental direction within Nebius AI, aiming to provide an end-to-end platform solution for all kinds of BigData and AI challenges, including data preparation, distributed training, offline model inference, and more. 

We have already secured several clients both within and outside Nebius AI. Some of these clients use TractoAI as their primary data lake or as a foundation for large language model (LLM) pre-training. 

Our Vision: We aim to transform YTsaurus into a leading BigData and ML product, competing with top cloud-like Data/AI platforms such as Databricks, Snowflake, VastData, MosaicML, and corresponding products from the hyperscaler clouds like AWS, Azure, or GCP. Our challenge lies in converting powerful technology into a successful product by defining the target audience, and expanding our existing client base. 

Role Overview: In this role, you will work on improving the YTsaurus core for the needs of our TractoAI platform. With us, you will work on performance and reliability of a large-scale distributed system. Your job responsibilities will include designing new features and microservices of the system, investigating performance issues of user workloads and collaborating with our SRE team to make our system stable. 

Our team is a small group of core YTsaurus developers with many years of experience in building distributed systems, standing behind the 100k-server-scale fleets of BigTech companies. We work on system architecture, software engineering and operations, making sure our code is not only committed and well-tested, but also works in production without disruptions. Our team is small, and we try to move fast to stay at the forefront of modern technologies. 

Representative projects: 

• Support S3 on a par with physical disks as storage for TractoAI 

• Implement fair strategy of disk throughput distribution to share IO resources between concurrent computations of different users on the cluster. 

• Make modern hardware like high-end GPUs or InfiniBand connectors work inside user job VMs and collect statistics about their utilization 

• Improve TractoAI scheduler to make splitting the input data to the jobs in MapReduce operation more uniform by using various metadata 

• Improve TractoAI IO engines to squeeze maximum out of the HDDs and NVMe SSDs. 

What We Offer: 

• A dynamic and collaborative work environment. 

• Opportunity to work on cutting-edge technology. 

• Flexible working conditions. 

• Competitive salary and benefits. 

We expect you to have: 

• Have 5+ years of experience as a software engineer 

• Have an experience with C++ 

• Have an experience with concurrency 

• Are result-oriented and ready to push a feature from the idea to the client adoption 

• Ready to dig into unknown areas, including HPC or GPU computing 

• Ready to occasionally write code in Go/Python 

It will be an added bonus if you have: 

• Distributed systems design 

• Understanding how databases work 

• OS internals 

• Performance engineering 

• Maintaining large-scale stateful system 

We’re growing and expanding our products every day. If you’re up to the challenge and are excited about AI and ML as much as we are, join us!