ResponsibilitiesYour mission is to own every dataset end to end - from discovering the source and securing access, to writing the pipelines that ingest it, to guaranteeing it enters training clean, standardized, and correct.
- Research and source new modalities of multimodal physical data (e.g. sparse sensors, point clouds, hyperspectral imagery, radar), and secure access through partnerships, vendors, and public archives
- Build petabyte-scale data pipelines (e.g. Apache Spark) that ingest each source into our storage in standardized, training-ready form, across both batch and streaming - including the orchestration, storage, and monitoring they need where shared platform infrastructure doesn't yet exist
- Develop quality metrics that measure coverage, correctness, and consistency across sources - and catch the subtle inconsistencies (sensor bias, drift, processing artifacts) that silently degrade models
- Design and implement automated QA checks that continuously measure and monitor data quality over time, and own the verdicts they produce
- Write technical requirements and provide actionable feedback to external data vendors and partners
- Collaborate with researchers to validate that new and improved datasets translate into model performance
What we're looking forWe value a relentless approach to problem-solving, rapid execution, and the ability to quickly learn in unfamiliar domains.
- Demonstrated experience building large-scale data pipelines, QA systems, or evaluation workflows (e.g. Spark, Ray, Beam)
- Detail-oriented in identifying subtle data inconsistencies and issues that could affect quality, with the ability to understand how quality impacts model performance
- Comfortable going deep on unfamiliar source material - reading format specifications, sensor documentation, and vendor manuals to get ingestion exactly right
- Experience working with external data vendors and partners, from technical evaluation to ongoing feedback
- Owns deliverables end-to-end, from collecting and translating requirements to autonomously driving execution
