RHC
Menu

← All projects

MLOps

modelcanary

Progressive delivery for ML models: a FastAPI gateway serves ONNX models in shadow and then canary stages. Sequential tests decide automatic promotion or rollback, and every decision goes to a tamper-evident audit trail.

modelcanary — Concept render
Concept render

Drawing

Diagram: a model moves from shadow to canary to promotion, with a rollback boundary

Highlights

  • Error rate, latency and a business metric are monitored with Wald SPRT and always-valid confidence sequences.
  • SHA-256 hash-chained audit log detects edits, deletions and reordering.
  • Evaluated against a ground-truth simulator over thousands of held-out seeds.