Skip to main content

AI Dataschema

A unified, production-grade data layer for AI & ML workloads.

Python Internal Package

AI Dataschema standardizes how you store, version, query, join, batch, stream, synchronize, and serve multimodal data — staying open, Arrow‑native, and framework‑agnostic.

Key capabilities:

  • Unified schema and metadata across modalities
  • Versioned datasets and their stored data with reproducible rollbacks
  • High-throughput batch/split readers and optional stream synchronization
  • Flexible decoding and a pluggable converter pattern
  • Seamless training integration with PyTorch datasets and dataloaders
  • Extensible design for new data types and pipelines

Installation​

Access requirement

You need read access to the internal Nexus repository to install this package. Contact alexander.ra.ross@bmw.de to request for access.

Supported Python: 3.10 – 3.12.

Install from internal Nexus (virtual environment recommended):

pip install --index-url https://nexus.bmwgroup.net/repository/pypi_hosted/simple --extra-index-url https://pypi.org/simple AIdataschema

Optional framework extras (requires Python 3.11 – 3.12):

# PyTorch only
pip install --index-url https://nexus.bmwgroup.net/repository/pypi_hosted/simple --extra-index-url https://pypi.org/simple AIdataschema[pytorch]

# PyTorch + TensorFlow
pip install --index-url https://nexus.bmwgroup.net/repository/pypi_hosted/simple --extra-index-url https://pypi.org/simple AIdataschema[pytorch,tensorflow]

Further Resources​

For full documentation including the user manuals, tutorials, and API reference, visit the AI Dataschema Documentation Portal.

Guides​

Click the card below for a hands-on look at basic usage of AI Dataschema.