Unlike traditional ML (batch training → deploy → retrain):
- Learns from every event incrementally
- Adapts to pattern shifts in ~2 minutes vs 3+ days
- No retraining pipeline needed
Demo: `docker compose up`
Watch the fraud detection model adapt automatically when fraudster tactics change at transaction 500.
dcris19710101•1d ago
Tech stack: Apache Kafka (KRaft), River (online ML), Hoeffding Trees, Streamlit
Unlike traditional ML (batch training → deploy → retrain): - Learns from every event incrementally - Adapts to pattern shifts in ~2 minutes vs 3+ days - No retraining pipeline needed
Demo: `docker compose up`
Watch the fraud detection model adapt automatically when fraudster tactics change at transaction 500.
Detailed writeup: https://medium.com/@dcris19740101/announcing-software-4-0-wh...
This pattern (continuous learning from events) is how Netflix, Uber, LinkedIn already work.
Feedback welcome!