AI/ML Systems
Operational ML systems connect data hygiene, model evaluation, delivery interfaces, and human oversight. This section documents public portfolio architecture for ERP-adjacent prediction workloads—without overstating production deployment.
Portfolio stance
| Claim type | Standard |
|---|---|
| Code structure in a public repo | OK when README and releases support it |
| Offline model metrics on synthetic/sample data | OK when labeled as such |
| Live production deployment | Only with explicit public evidence |
| Client or employer outcomes | Not in this public portfolio |
Featured public repository
ERP AI Delay Risk
Portfolio machine-learning architecture exploring operational delay-risk signals from ERP-adjacent features—feature engineering, model evaluation, API delivery patterns, batch and real-time scoring concepts, and reporting hooks.
See ERP Delay Risk for an honest capability and status breakdown.
ML lifecycle (reference model)
Problem framing → Label definition → Feature pipeline → Train/evaluate
→ Bias/limitation review → Packaging → Delivery (batch/API)
→ Monitor drift → Human override → Evidence archive
Each stage should produce artifacts a reviewer can inspect without private data access.
Prerequisites from ERP transformation
Models built on ERP movement, order, or shipment history inherit:
- Master data quality (ERP Data Quality)
- Warehouse timing integrity (Supply Chain and Warehouse)
- Costing consistency where labels depend on fulfillment state (FIFO and Inventory Costing)
Run the AI Readiness Diagnostic before treating ML as the next step.
Governance
- Human authority for consequential decisions
- No automated approval based solely on model score
- Explicit documentation of label leakage and exclusion rules
- Public repos must not contain private extracts