About the role
<div class="content-intro"><p><strong>About Nebius:</strong></p> <p>Nebius is leading a new era in cloud infrastructure for the global AI economy. We are building a full-stack AI cloud platform that supports developers and enterprises from data and model training through to production deployment, without the cost and complexity of building large in-house AI/ML infrastructure.</p> <p>Built by engineers, for engineers. From large-scale GPU orchestration to inference optimization, we own the hard problems across compute, storage, networking and applied AI.</p> <p>Listed on Nasdaq (NBIS) and headquartered in Amsterdam, we have a global footprint with R&D hubs across Europe, the UK, North America and Israel. Our team of 1,500+ includes hundreds of engineers with deep expertise across hardware, software and AI R&D.</p></div><h4 id="The-role" data-local-id="c41794725a88" data-renderer-start-pos="1">The role</h4> <p>#LI-MM1</p> <p data-renderer-start-pos="11" data-local-id="b628b6030a86">We’re looking for a <strong data-renderer-mark="true">Software Engineer with strong C++ expertise</strong> to join the team building and operating <strong data-renderer-mark="true">Nebius Data Platform</strong> — a distributed storage and a processing platform that acts as the company’s “source of truth” and the backbone of many internal (and some external) products.</p> <p data-renderer-start-pos="292" data-local-id="de1e1b789ce4">Nebius Data Platform is a <strong data-renderer-mark="true">single multi-tenant ecosystem based on YTsaurus</strong> — instead of running separate HDFS/Kafka/HBase-style systems, we provide storage, compute, and analytics capabilities inside one platform.</p> <p data-renderer-start-pos="506" data-local-id="bcbb3787b5d5">Built on top of the open-source YTsaurus ecosystem, we run and extend our own Nebius distribution and develop significant in-house functionality (core and platform-level). We can design, implement, and roll out features end-to-end on our clusters without waiting for upstream approvals and contribute upstream when it makes sense.</p> <p data-renderer-start-pos="838" data-local-id="7544511b100f">At scale today, this includes<strong data-renderer-mark="true">~500 servers, ~20k CPU cores</strong> and <strong data-renderer-mark="true">~10 PB of compressed</strong> <strong data-renderer-mark="true">data</strong> in our largest production cluster, supporting workloads ranging from business-critical pipelines and financial transactions to large-scale ML/LLM training datasets and compute.</p> <h4 id="What’s-inside-the-platform" data-local-id="2100fd8e190a" data-renderer-start-pos="1105">What’s inside the platform</h4> <p data-renderer-start-pos="1133" data-local-id="76f8ed17de21">You’ll work on a system that includes (and ties together):</p> <ul> <li data-renderer-start-pos="1195" data-local-id="6ddeb26cab43"><strong data-renderer-mark="true">Distributed Storage (Cypress)</strong>: transactional semantics, tiered storage, erasure coding, replication, and strong reliability expectations.</li> <li data-renderer-start-pos="1336" data-local-id="13562646bfb3"><strong data-renderer-mark="true">Compute & ETL</strong>: a cluster-wide job scheduler (tens of thousands of cores), MapReduce, <strong data-renderer-mark="true">YQL</strong> for SQL-like data processing, and <strong data-renderer-mark="true">SPYT (Spark over YTsaurus)</strong> for modern data engineering.</li> <li data-renderer-start-pos="1518" data-local-id="97897e1cbf8b"><strong data-renderer-mark="true">Interactive analytics (CHYT)</strong>: ClickHouse® instances spun up directly on compute nodes for fast SQL over data in-place.</li> <li data-renderer-start-pos="1640" data-local-id="02d7848ef2e9"><strong data-renderer-mark="true">Dynamic Tables</strong>: low-latency NoSQL KV with distributed ACID transactions for OLTP-style workloads and feature stores.</li> <li data-renderer-start-pos="1760" data-local-id="bcaf763a8a8a"><strong data-renderer-mark="true">Orchestracto</strong>: workflow orchestration deeply integrated with the platform (Airflow-like, but platform-native).</li> </ul> <h4 id="What-you’ll-do" data-local-id="ef3970644491" data-renderer-start-pos="1873">What you’ll do</h4> <p data-renderer-start-pos="1889" data-local-id="76f1e15bff02">We’re looking for engineers who combine strong systems skills with <strong data-renderer-mark="true">product sense</strong>: understanding who uses the platform, why certain capabilities matter, and making pragmatic trade-offs to maximize impact. On our team, engineering work is expected to be connected to real users and outcomes — you’ll regularly align with internal stakeholders, clarify requirements, and help drive prioritization.</p> <p data-renderer-start-pos="2285" data-local-id="55415fa782de">In this role, you will:</p> <ul> <li data-renderer-start-pos="2312" data-local-id="976b454233f2"><strong data-renderer-mark="true">Design and implement new functionality in YTsaurus core</strong> (C++) with production reliability in mind.</li> <li data-renderer-start-pos="2414" data-local-id="2003d4e11122"><strong data-renderer-mark="true">Build and evolve platform-level capabilities:</strong> platform architecture and operating model—multi-cluster growth, shared primitives, and a consistent experience that scales with new teams and use cases.</li> <li data-renderer-start-pos="2616" data-local-id="fd06f5a1ee0e">Improve <strong data-renderer-mark="true">end-to-end platform experience</strong> for internal (and external-facing) users: APIs, guardrails, debugging workflows, and automation.</li> <li data-renderer-start-pos="2755" data-local-id="c2f3a7ce1432">Own production quality: <strong data-renderer-mark="true">incident response / on-call rotation</strong>, root cause analysis, and turning learnings into durable fixes.</li> </ul> <h4 id="Example-projects" data-local-id="e54b80d154d0" data-renderer-start-pos="2883">Example projects</h4> <ul> <li data-renderer-start-pos="2903" data-local-id="67c11986cf10">Roll out sharded YTsaurus masters (incl. Kubernetes operator support) and build automatic balancing of metadata across master cells (consensus groups) to remove control-plane bottlenecks and <strong data-renderer-mark="true">unlock 10–100x cluster growth</strong>.</li> <li data-renderer-start-pos="3128" data-local-id="61b805a294c7"><strong data-renderer-mark="true">Make CHYT interactive SQL faster and more predictable at high load</strong> via performance work like data-skipping / min-max-style indexes and improved execution introspection.</li> <li data-renderer-start-pos="3300" data-local-id="8dcfe4b24784">Turn <strong data-renderer-mark="true">Orchestracto into a platform product </strong>by defining the building blocks, developer experience, and governance for how teams create and share workflows.</li> <li data-renderer-start-pos="3457" data-local-id="b8561c1450f0"><strong data-renderer-mark="true">Scale and harden Parquet-on-S3 for native YTsaurus workloads</strong> by tackling replication/movement, consistent lifecycle semantics, and master-server metadata optimizations for performance and reliability.</li> <li data-renderer-start-pos="3661" data-local-id="55b0019ca4db">Design and ship <strong data-renderer-mark="true">complete, trustworthy audit trails for data changes</strong> (who/what/when) across heterogeneous storage and compute paths.</li> </ul> <h4 id="Tech-stack" data-local-id="5ea174565779" data-renderer-start-pos="3796">Tech stack</h4> <ul> <li data-renderer-start-pos="3810" data-local-id="7558ae4388dc">Core: <strong data-renderer-mark="true">modern C++</strong> (C++20, async + multithreaded primitives)</li> <li data-renderer-start-pos="3872" data-local-id="44f7dddeadcc">Services & tooling: <strong data-renderer-mark="true">Go</strong> and <strong data-renderer-mark="true">Python</strong> (microservices, utilities, integration tests)</li> </ul> <h4 id="What-we-expect" data-local-id="47ea8a0da9c3"…