Software Engineer, ML Data Systems

Our mission is to automate coding. The first step in our journey is to build the best tool for professional programmers, using a combination of inventive research, design, and engineering. Our organization is very flat, and our team is small and talent dense. We particularly like people who are truth-seeking, passionate, and creative. We enjoy spirited debate, crazy ideas, and shipping code. About the Role Cursor ships daily. Every release leaves signals behind: telemetry, prompts, completions, agent runs, sessions. Those signals power model improvement, evals, and experimentation. Data infrastructure is what turns them into something teams can trust. A lot of systems here started simple so we could move fast. Over time, the constraints change and the “good enough” version becomes the bottleneck. This role owns the full ladder: patch what should be patched, redesign what should be redesigned, ship the replacement, and operate it. Privacy guarantees are part of correctness. What we can retain and use depends on Privacy Mode and org configuration, and getting that wrong breaks a product promise. We choose work by business impact: what blocks product and model teams today, and what will block them next month. Sample projects include... - A core pipeline started as a pragmatic reuse of infrastructure built for something else. It works, but it cannot guarantee properties downstream consumers now need (for example, point-in-time consistency). You design and ship the replacement while keeping the existing system running. - A new product surface ships without instrumentation. You talk to the team, define what needs to be captured, and wire it through before the absence becomes anyone else’s problem. - Eval coverage drops. You trace it to an instrumentation gap introduced weeks ago by a product change nobody flagged. You fix the gap, add a contract so it cannot recur, and ship the dashboard that would have caught it earlier. - Multiple consumers depend on overlapping data. You design schema evolution and validation so changes in one place do not silently degrade the others. - Storage costs rise faster than usage. You decide what is worth keeping, implement retention and compression, and delete what is not. What we're looking for We’re looking for someone who has built real systems at scale and cares about correctness, cost, and ergonomics. Strong signals include: - Deep experience with Spark (Databricks or open-source Spark both count) - Production experience with Ray Data - Hands-on ownership of large data pipelines and storage systems - Comfort debugging performance issues across client instrumentation, streaming, storage, and model-facing workflows, as well as, compute, storage, and networking layers - Clear thinking about data modeling and long-term maintainability - You have good judgment about when to patch and when to rebuild Nice to have - Experience running or scaling ClickHouse - Familiarity with dbt, Dagster, or similar orchestration and modeling tools We're in-person with cozy offices in North Beach, San Francisco and Manhattan, New York, replete with well-stocked libraries. Applying If there appears to be a fit, we'll reach to schedule 2-3 short technicals. After, we'll schedule an onsite in our office, where you'll work on a small project, discuss ideas, and meet the team. #LI-DNI

Strong experience building real-scale data systems with attention to correctness, cost, and ergonomics. Deep experience with Spark (Databricks or open-source Spark). Production experience with Ray Data. Hands-on ownership of large data pipelines and storage systems. Ability to debug performance issues across client instrumentation, streaming, storage, and model-facing workflows, as well as compute, storage, and networking layers. Clear thinking about data modeling and long-term maintainability. Good judgment about when to patch and when to rebuild. Nice-to-have: experience running or scaling ClickHouse; familiarity with dbt, Dagster, or similar orchestration and modeling tools.

Own the full data infrastructure lifecycle for internal signals (telemetry, prompts, completions, agent runs, sessions): patch where needed, redesign when required, ship replacements, and operate the system. Ensure privacy guarantees are upheld as part of correctness. identify and address blockers for product and model teams, today and in the near term. Design and ship data surfaces and instrumentation as needed, maintain data schemas, and ensure reliable, performant data flows across instrumentation, streaming, storage, and model-facing workflows. Collaborate with teams to define what needs to be captured, wire it through the system, and build dashboards or instrumentation to monitor health and quality.