NVIDIA · nvidia.wd5.myworkdayjobs.com · checked today
Senior Physical AI Data Engineer
NVIDIA is building the next generation of AI and advanced data technologies for Physical AI. Our MLOps Ground Truth (GT) team develops large-scale data infrastructure, tooling, AI-based GT generation systems, DNN auto labeling, VLM labeling, and agentic workflows for global automotive and robotics partners. We are looking for a Senior Physical AI Data Engineer to architect, build, and scale this global data ecosystem.
Skills, with evidence
- Python
Strong programming skills in Python and C++, with experience designing maintainable, well-tested software and troubleshooting complex production systems.
must have - Data quality
Build observability and data quality frameworks covering pipeline health, job execution, labeling quality, data lineage, and GT release readiness.
must have - Machine learning
A solid understanding of computer vision, machine learning, VLMs, and AI agent development, including evaluation and production integration.
must have · not practised here - Data modelling
Experience building reliable data pipelines with data validation, versioning, lineage, and monitoring.
must have - Governance & security
Build observability and data quality frameworks covering pipeline health, job execution, labeling quality, data lineage, and GT release readiness.
not practised here - AWS
Experience delivering multi-region data platforms with access controls, data governance, and regional data-handling requirements.
not practised here - Cost & performance
Optimize distributed processing and GPU inference workloads to improve throughput, resource utilization, cost efficiency, and release turnaround time.
- Failure handling
Experience building reliable data pipelines with data validation, versioning, lineage, and monitoring.
Your plan
- The SQL screen: correct, then fast≈ 2 h
Data quality
- Late arriving event auditIntermediate
- Order anomaly auditIntermediate
- Premium and free listening by countryIntermediate
- Interview audit union capstoneAdvanced
- Airbnb: converting on a day with no published rateAdvanced
- Python: the data-wrangling round≈ 3 h
Python
- Diff two snapshots of a tableIntermediate
- Explode an array column into rowsIntermediate
- Flatten nested event payloadsIntermediate
- Keep only the latest row for each eventIntermediate
- Pivot a long metrics table to wideIntermediate
- Data modelling: the round most people fail≈ 3 h
Data modelling
- Addresses that stay true to the pastIntermediate
- Seat holds and the release-night raceIntermediate
- Subscription warehouse grainIntermediate
- Campaign efficiencyIntermediate
- Catalogue: products, variants and sellersIntermediate
- Pipeline design: safe to run twice≈ 4 h
Failure handling
- Is this change safe?Intermediate
- The source will not let youIntermediate
- Changing a pipeline that’s already runningIntermediate
- SLA-aware alerting flowIntermediate
- The table everyone depends onIntermediate
- Spark: read the plan Spark actually ran≈ 2 h
Cost & performance
- applyInPandas per country: what Spark ships to Python, and the native rewriteAdvanced
- Filters you wrote in the wrong place: where Catalyst moves themAdvanced
- snappy, gzip or zstd: measure the Parquet codec trade-off yourselfAdvanced
- Find the four suspects in a slow pipeline (one of them is innocent)Advanced
- A filter after a window function: pushed down or not?Advanced
- Say it out loud≈ 1 h
25 drills · Intermediate + Advanced≈ 14 hours
Not covered by the plan: Machine learning, Governance & security, AWS.
Readiness
Counted from drills you have completed anywhere on D8LooP.
leaves in 13 dremoved the moment NVIDIA closes it