Structure profiling
Map file organization, fields, modalities, topics, and relationships.
Data Engineering
From raw embodied data to training-ready datasets
KeenTruth focuses on customer-supplied data before it enters model training: governance, annotation, quality assurance, and standardized delivery.
We do not provide data collection or scene construction. Customers provide raw data and training requirements; KeenTruth turns them into organized, semantically consistent, and verifiable training data.
01
Data formats should not limit data engineering
We support video, ROS bag, MCAP, HDF5, and adaptable workflows for other heterogeneous or custom formats.
- Visual, state, action, control, and sensor streams
- Structure and governance across embodied-data topics
- Timestamp, field-definition, and relationship checks
- Custom schemas and output adaptation
02
Data governance scope
Completeness checks
Detect corrupted files, missing topics, interruptions, and duplicate data.
Anomaly filtering
Locate invalid states, failed tasks, unusable operations, and out-of-scope data.
Association and structuring
Organize observations, states, actions, controls, and task information.
03
Standard workflow
- 01Align model objectives, data scope, and acceptance criteria
- 02Profile formats, structures, and topics
- 03Run completeness checks and issue reporting
- 04Define governance rules, labels, and exception handling
- 05Calibrate through a representative pilot
- 06Scale governance, annotation, review, correction, and acceptance
04
Deliverables
- Governed valid data and structured annotations
- Task and action hierarchies, object boxes, and hand keypoints
- Anomaly and issue reports
- Quality statistics, acceptance results, versions, and batch records
FAQ
Frequently asked questions
Does KeenTruth collect data?
No. KeenTruth focuses on governance, annotation, QA, and delivery for data customers already have.
Do we need to provide an annotation platform?
No. KeenTruth runs production and quality control on its proprietary platform and delivers the agreed output.
Do you only support ROS bag, MCAP, and HDF5?
No. We evaluate other heterogeneous or custom formats using sample data, field documentation, and delivery goals.
Can KeenTruth provide QA without annotation?
Yes. We can inspect raw data, third-party annotations, or data prepared for delivery.
Contact
Move raw embodied data into training faster.
Tell us about your data, annotation requirements, and quality expectations.