• Standalone
  • High-performance
  • Training sets for ML, NLP do not need to be reuseable
  • Interoperability is not required
  • Explanation support is not required