Essential session storage is required to sign in. Optional analytics remain off unless you allow them.
Build a production-shaped ML system that closes the loop. Include: (1) versioned data + a feature definition that serves the SAME features to training and inference (avoiding training-serving skew); (2) an orchestrated training pipeline (validate→train→evaluate→register) with experiment tracking and a model registry; (3) CI/CD/CT — CONTINUOUS TRAINING triggered by schedule AND by detected drift, re-running data/model gates before promotion; (4) a served model with a safe deployment strategy and rollback; (5) monitoring that distinguishes DATA DRIFT from CONCEPT DRIFT and feeds the retraining trigger; and (6) governance — lineage/reproducibility good enough to audit any prediction, plus a model card. Document the maturity level you achieved and the next step.
Orchestrated train→eval→register pipeline; Continuous Training triggered by schedule + drift, re-running quality gates before promotion; registry-driven.
Consistent features for training and serving (feature store or shared transform); data versioned; training-serving skew avoided.
Monitoring distinguishes data drift from concept drift and feeds the retraining trigger; safe deployment + rollback; feedback loop closed.
Auditable lineage/reproducibility for any prediction; a model card; stated maturity level and next step.
Use this README structure to make the work reviewable.
DevPath fetches a read-only, size-capped snapshot. Submitted code is never executed.
# An end-to-end MLOps system with continuous training ## Goal Build a production-shaped ML system that closes the loop. Include: (1) versioned data + a feature definition that serves the SAME features to training and inference (avoiding training-serving skew); (2) an orchestrated training pipeline (validate→train→evaluate→register) with experiment tracking and a model registry; (3) CI/CD/CT — CONTINUOUS TRAINING triggered by schedule AND by detected drift, re-running data/model gates before promotion; (4) a served model with a safe deployment strategy and rollback; (5) monitoring that distinguishes DATA DRIFT from CONCEPT DRIFT and feeds the retraining trigger; and (6) governance — lineage/reproducibility good enough to audit any prediction, plus a model card. Document the maturity level you achieved and the next step. ## Acceptance criteria - [ ] Pipeline & CT: Orchestrated train→eval→register pipeline; Continuous Training triggered by schedule + drift, re-running quality gates before promotion; registry-driven. - [ ] Features & no skew: Consistent features for training and serving (feature store or shared transform); data versioned; training-serving skew avoided. - [ ] Monitoring & loop: Monitoring distinguishes data drift from concept drift and feeds the retraining trigger; safe deployment + rollback; feedback loop closed. - [ ] Governance: Auditable lineage/reproducibility for any prediction; a model card; stated maturity level and next step. ## Verification - [ ] Document installation and run commands. - [ ] Record automated checks and their exact commands. - [ ] Add representative output, screenshots, or a short demo where useful. - [ ] Confirm failure paths and known constraints. ## Decision log ### Decision title - Context: - Choice: - Alternatives considered: - Tradeoffs: ## Evidence - Link each rubric criterion to the file, test, or artifact that demonstrates it. ## Known limitations - List what is intentionally out of scope and what should be improved next.