• What you'll doOwn the flywheel — logging correctness across video, audio, events, and telemetry (timestamp discipline included); the datalake with its ingest, storage, and labeling tooling
• Make data self-serve — ML engineers assemble datasets without a human intermediary; every deployed model traces back to its exact training data
• Guard the privacy promise — verify anonymization at ingest, don't assume it
• Feed the research — build the export/encoding pipeline for our egocentric robotics dataset, the foundation of our robot-learning thesis
• What you getSole ownership of the foundation two missions stand on — client delivery today, robot-learning research tomorrow
• Schema and lineage decisions that compound for years, made by you, early
• Data most teams never see: real egocentric multimodal recordings from factory floors

Where you'll be in 12 monthsA field recording becomes ready for labeling in hours. Dataset assembly needs nobody's help. Nothing gets lost, and every model that passed client acceptance has full, queryable lineage. That's what we'll build together — and it will be yours.

• Who you areYou treat data quality as the product, not a chore
• Privacy by default — you verify, you don't assume
• You take pride in invisible infrastructure others build on without thinking about it
• You anticipate what ML consumers need before they file the ticket
• You can debug a pipeline across five systems without owning any of them fully

• Your experienceMust have:Strong Python
• Production experience with relational and non-relational databases
• Object storage and data pipelines at real scale
• Data serialization formats and their evolution (e.g., protobuf, Avro)
• Schema design and migrations (e.g. Alembic)
• Practical video/audio data handling: codecs, segmenting, timestamps
• Ways to stand out:A system language (C++, Rust) alongside Python
• ML data labeling and curation tooling (e.g. Label Studio, FiftyOne)
• Edge-to-cloud sync
•