COMPEL Certification Body of Knowledge — Module 4.2: Framework Interoperability and Integration Architecture
Article 7 of 10
DevOps transformed software delivery by unifying development and operations into a continuous delivery pipeline. MLOps extends this paradigm to machine learning — creating the engineering practices, tooling, and cultural norms that enable AI models to be developed, deployed, monitored, and maintained with the same velocity and reliability that DevOps brought to traditional software. The EATP Lead must integrate COMPEL's transformation methodology with DevOps/MLOps engineering practices, ensuring that the strategic transformation agenda can be executed at the pace that modern engineering enables.
Understanding the DevOps/MLOps Landscape
DevOps Foundations
DevOps is built on several core practices that have become standard in modern software engineering:
- Continuous Integration (CI): Automated building and testing of code changes as they are committed
- Continuous Delivery (CD): Automated deployment of validated changes to production environments
- Infrastructure as Code (IaC): Managing infrastructure through version-controlled configuration files
- Monitoring and Observability: Comprehensive instrumentation that provides visibility into system behavior
- Incident Response: Structured processes for detecting, responding to, and learning from production incidents
- Site Reliability Engineering (SRE): Applying software engineering principles to operations, with error budgets and SLOs
MLOps Extensions
MLOps extends DevOps to address the distinctive characteristics of machine learning systems:
- Data Versioning: Tracking and managing the datasets used for model training, with the same rigor that source control brings to code
- Experiment Tracking: Recording the parameters, metrics, and artifacts of every model training experiment
- Model Registry: A centralized repository of trained models, with versioning, metadata, and approval workflows
- Feature Stores: Centralized repositories of engineered features that can be shared across models and teams
- Model Serving: Infrastructure for deploying models to production environments — batch, real-time, and edge
- Model Monitoring: Continuous monitoring of deployed models for performance degradation, data drift, and concept drift
- Pipeline Orchestration: Managing the end-to-end ML pipeline — data ingestion, feature engineering, model training, evaluation, deployment, and monitoring — as an automated, reproducible workflow
The Integration Architecture
COMPEL Technology Domains and DevOps/MLOps Maturity
COMPEL's technology maturity domains (Domains 10-13 in the 18-domain maturity model) directly map to DevOps/MLOps capabilities. The EATP Lead uses this mapping to assess and develop the organization's engineering velocity:
| COMPEL Technology Domain | DevOps/MLOps Capability | Maturity Indicator |
|---|---|---|
| Domain 10: AI/ML Platforms | ML platform maturity | Self-service vs. manual provisioning |
| Domain 11: Data Infrastructure | Data pipeline maturity | Automated, versioned pipelines vs. ad hoc data handling |
| Domain 12: Integration Architecture | API and deployment maturity | Continuous deployment vs. manual deployment |
| Domain 13: Security and Infrastructure | Security automation maturity | Automated security scanning vs. manual review |
Pipeline-Stage Integration
The EATP Lead maps COMPEL lifecycle stages to DevOps/MLOps pipeline stages, ensuring that transformation governance integrates with engineering execution:
Calibrate Stage — Assessment Pipeline: Automated collection of engineering metrics — deployment frequency, lead time for changes, change failure rate, mean time to recovery (the DORA metrics) — feeds into COMPEL's maturity assessment. The assessment pipeline provides objective, real-time data on the organization's engineering capability maturity.
Organize Stage — Platform Provisioning: The transformation organization design includes DevOps/MLOps platform teams responsible for building and maintaining the engineering infrastructure. COMPEL's organizational design frameworks from Module 3.2 inform the structure of platform engineering teams.
Model Stage — Architecture and Design: The target state design includes the target MLOps architecture — the platforms, pipelines, and practices that the organization needs to achieve its AI maturity goals. This architecture is expressed using the patterns from Module 4.2, Article 4: COMPEL and TOGAF — Enterprise Architecture Integration.
Produce Stage — Continuous Delivery: AI capabilities are developed and deployed through MLOps pipelines. COMPEL's execution governance ensures that delivery maintains quality standards without impeding engineering velocity.
Evaluate Stage — Automated Assessment: Model performance, data quality, fairness metrics, and operational health are automatically evaluated through monitoring pipelines, providing continuous input to COMPEL's maturity assessment.
Learn Stage — Continuous Improvement: Engineering retrospectives, post-incident reviews, and experiment results feed organizational learning and drive pipeline improvements.
MLOps Maturity Model
The EATP Lead applies an MLOps maturity model that aligns with COMPEL's five maturity levels:
Level 1 — Ad Hoc
Models are developed in notebooks and deployed manually. No version control for data or models. No automated testing. No monitoring. Each deployment is a unique, manual effort.
Level 2 — Managed
Basic version control for code. Manual but documented deployment processes. Some monitoring of deployed models. Individual teams have their own tools and practices.
Level 3 — Defined
Standardized ML pipelines with automated training and deployment. Model registry for version management. Automated testing for model performance. Centralized monitoring. Feature stores for shared features.
Level 4 — Measured
Comprehensive automated pipelines from data ingestion through model deployment and monitoring. Automated retraining triggered by performance degradation. Advanced monitoring including fairness, explainability, and drift detection. DORA metrics tracked and optimized.
Level 5 — Optimized
Fully automated, self-healing ML pipelines. Automated model selection and hyperparameter optimization. Real-time A/B testing and canary deployments. Automated compliance and governance checks integrated into the pipeline. Continuous optimization of the pipeline itself.
Governance Without Friction
The central tension in DevOps/MLOps integration is governance. COMPEL requires governance — stage gates, reviews, approvals, compliance checks. DevOps/MLOps values velocity — fast deployments, automated processes, minimal manual intervention. The EATP Lead must resolve this tension by embedding governance into the pipeline itself, not bolting it on as a separate process.
Automated Governance Gates: Compliance checks, security scans, fairness assessments, and documentation requirements are implemented as automated pipeline stages. If the automated checks pass, the deployment proceeds without manual intervention. If they fail, the pipeline stops and the appropriate review process is triggered.
Policy as Code: Governance policies — model performance thresholds, data quality requirements, bias tolerance levels, documentation standards — are expressed as code that the pipeline evaluates automatically. This makes governance transparent, testable, and version-controlled.
Audit Trail Automation: Every pipeline execution automatically generates an audit trail — who changed what, when, why, with what approval, and what the results were. This satisfies governance requirements for accountability and traceability without requiring manual documentation.
Risk-Based Approval Tiers: Not every deployment requires the same level of governance scrutiny. The EATP Lead establishes deployment tiers based on risk — low-risk deployments (minor model updates in non-critical applications) proceed automatically; high-risk deployments (new models in regulated domains) require human review. This tiered approach applies governance proportionate to risk, maximizing both velocity and control.
Platform Engineering and COMPEL
The EATP Lead recognizes that MLOps capability is ultimately a platform engineering challenge. The organization needs an internal AI platform that provides data scientists and ML engineers with self-service access to data, compute, training infrastructure, deployment targets, and monitoring tools — all governed by the policies and standards that COMPEL's governance framework establishes.
This platform engineering perspective connects to Module 4.4: Enterprise AI Operating Model Design, where the EATP Lead designs the organizational structures and capabilities required to build and sustain the AI platform at enterprise scale.
The next article, Module 4.2, Article 8: COMPEL and COBIT — IT Governance Convergence, addresses the integration with COBIT, the framework that provides the overarching IT governance structure within which all other frameworks operate.
© FlowRidge.io — COMPEL AI Transformation Methodology. All rights reserved.