Back to browse
GitHub Repository

Reliability layer for delayed-label ML under distribution shift

1 starsPython

A controller to automatically mitigate model drift

by pberlizov·Jun 9, 2026·1 point·1 comment

AI Analysis

●●SolidNiche GemBig Brain

Handles 12-week label delays in fraud models without scheduled retrains.

Strengths
  • Distinguishes shifts needing intervention from those that self-resolve.
  • Learns from delayed feedback like chargebacks and disputes.
  • Benchmarked across 3 public fraud streams with measurable risk reduction.
Weaknesses
  • Narrow applicability to delayed-label scenarios only.
  • Requires PyTorch integration for deployment.
Category
Target Audience

ML engineers in fraud, AML, delayed-label scenarios

Similar To

Evidently AI · Arize · WhyLabs

Similar Projects