ConceptNet 4-layer enterprise voice intent classifier, 98.6% accuracy, 9 languages. Looking for feedback.
Hi Hugging Face community 👋🏿
I've just published ConceptNet — a fine-tuned DistilBERT multilingual model for enterprise voice intent classification.
What it does:
Classifies enterprise voice commands into exactly 4 intent layers:
L1 Basic — "Do X" — immediate execution
L2 Context-Aware — "Do X when Y" — conditional
L3 Predictive — "Do X before Y" — proactive
L4 Autonomous — "Do X always" — persistent agent
The numbers:
Fast-path classifier: 83% accuracy, <5ms
Neural model: 98.6% accuracy, <100ms
Trained on 730 examples across 9 languages
Constrained grammar output — exactly 4 labels, no hallucination
Try it:
Live sandbox: conceptnet.co.uk/sandbox/
Model: huggingface.co/conceptnetUk/intent-classifier
GitHub: github.com/wushu75/ConceptNet
Would love feedback from the community — especially anyone working on enterprise NLP, intent classification, or multilingual models.
Also looking for enterprise teams worldwide to pilot free. If your team does repetitive voice or text workflows — sales, legal, ops, government — try the sandbox and send me your thoughts.