I built a text classifier that can learn new categories on the fly without full retraining – solving a practical problem I kept hitting in production systems. Try it out:

from adaptive_classifier import AdaptiveClassifier

# Start with basic sentiment classes classifier = AdaptiveClassifier("distilbert-base-uncased") classifier.add_examples( ["Great product!", "Terrible service"], ["positive", "negative"] )

# Later add a completely new class without retraining classifier.add_examples( ["404 error", "API timeout"], ["technical", "technical"] )

# It maintains accuracy on both old and new classes print(classifier.predict("The API is down")) # [('technical', 0.82), ...] print(classifier.predict("Best purchase ever")) # [('positive', 0.89), ...]

Technical implementation:

- Uses Elastic Weight Consolidation to prevent catastrophic forgetting when learning new classes

- Combines prototype memory with neural adaptation for efficient few-shot learning

- Memory-efficient: stores only representative examples using k-means clustering

- Fully deterministic results through careful seed management

- Built on PyTorch/HuggingFace, works with any transformer model

The key insight was combining EWC (usually used in reinforcement learning) with a prototype memory system. This lets the model learn new patterns while preserving its knowledge of existing categories.

Real-world use cases I've tested:

- Customer feedback classification where new issue types emerge over time

- Content moderation systems that need to handle new violation types

- Technical support routing where new product categories get added

pip install adaptive-classifier

GitHub: https://github.com/codelion/adaptive-classifier

Would love feedback from anyone who's dealt with similar challenges in production ML systems!