Applying Integrated Frameworks to Generative AI and Modern Data Architectures

This lesson equips data architects with the tools to extend their integrated framework stack—TOGAF, DAMA-DMBOK, and SAP Enterprise Architecture Framework—into cutting-edge areas like generative AI, Zero-ETL integration, and federated data sharing.

Watch the video for an overview.

Start with generative AI building blocks, which transform raw data into intelligent applications. Vector databases store high-dimensional embeddings—numerical representations of text, images, or other data—that enable semantic search and similarity matching. Retrieval-Augmented Generation (RAG) pipelines combine these vectors with large language models: relevant data is retrieved from the vector store, augmented into prompts, and generated into responses. Guardrails enforce safety through content filtering, toxicity detection, and compliance checks, while feedback loops capture user interactions to refine models and data over time.

Expressed as architecture patterns, these become standardized building blocks:

• A RAG Reference Pattern might include data ingestion, embedding generation, vector indexing, retrieval orchestration, and output validation.
• Feedback loops form a Model Observability Pattern with metrics for accuracy, drift, and bias.

Using TOGAF's content framework, these patterns populate the Architecture Repository as reusable assets, available across domains.

Integrating Generative AI into TOGAF begins in the Architecture Vision and Business Architecture phases, where requirements capture use cases like customer personalization or automated reporting. Principles emerge, such as Responsible AI with Human Oversight or Data Provenance for Explainability. In Data Architecture phases, target states define vector databases alongside traditional lakes and warehouses, with transition architectures sequencing pilots (e.g., single-domain RAG) to enterprise scale. Risk management in the Requirements Management and Governance phases addresses hallucinations, data leakage, and vendor lock-in.

DAMA-DMBOK complements this through its knowledge areas as illustrated in the figure:

Together, TOGAF provides the "what and when," while DAMA delivers the "how," ensuring generative AI is not experimental but enterprise-grade.

Next, trace the evolution from Extract, Transform, Load (ETL) to modern paradigms, which fundamentally reshape data architecture.

Traditional ETL centralized transformation in rigid pipelines often becoming a bottleneck for agility. Extract, Load, Transform (ELT) flips this, loading raw data into scalable cloud storage for on-demand processing. Streaming adds real-time capabilities via event-driven architectures, processing data as it arrives. Zero-ETL takes this further, eliminating transformation layers altogether through native platform integrations—like querying operational databases directly from analytical warehouses without copying data. Platform-level data sharing, enabled by lakehouses (e.g., Delta Lake) and data marketplaces, allows secure, governed access across organizations via catalogs, contracts, and fine-grained permissions. These shifts impact architecture by favoring logical over physical integration: data stays in place, accessed virtually, reducing costs and latency while amplifying governance needs.

DAMA-DMBOK operationalizes these patterns.

• Data Integration standards specify when to use federation, replication, or caching for movement-minimizing designs. This is guided by policies like "Minimize Data Mobility Unless Performance Dictates Otherwise."
• Data Sharing Agreements become formal artifacts under Data Governance, outlining SLAs, quality thresholds, and liability.
• Data Contracts—machine-readable schemas with quality rules, evolution policies, and access terms—emerge from Metadata and Data Modeling practices, ensuring interoperability.
• Access Control leverages Data Security domains, implementing role-based, attribute-based, and purpose-based mechanisms across domains and platforms.
• Interoperability standards from Documents and Data Operations prevent silos, mandating common formats and lineage tracking.

These DAMA outputs plug directly into TOGAF deliverables, such as data architecture diagrams showing virtualized access layers.

Finally, embed these patterns into governance and operating models for longevity. TOGAF's Architecture Governance provides the board structure—Architecture Boards review initiatives against principles and standards. DAMA-inspired Data Councils handle domain-specific decisions, with clear decision rights via RACI matrices:

• Data Owners approve products
• Stewards enforce quality
• Platforms provide infrastructure

KPIs measure success: generative AI accuracy (>95%), Zero-ETL uptime (99.9%), data contract compliance (100%), and sharing velocity (products/month).

Feedback loops close the circle: post-implementation reviews feed into Architecture Repositories, updating patterns and principles. SAP Enterprise Architect Framework tooling, like SAP LeanIX, visualizes these evolutions, tracking maturity from ad-hoc generative AI to federated mesh. This living model ensures the framework stack adapts quarterly, incorporating lessons from pilots and regulatory shifts.

By mastering this section, architects position their organizations at the forefront: generative AI drives value, Zero-ETL accelerates insights, sharing unlocks ecosystems, and governance sustains trust. The integrated frameworks turn complexity into competitive advantage, scalable from today’s proofs-of-concept to tomorrow’s standards.

This lesson explains how to use TOGAF, DAMA-DMBOK, and the SAP Enterprise Architecture Framework to manage generative AI, Zero-ETL, and data sharing.

• Technical components such as vector databases, Retrieval-Augmented Generation (RAG), and safety guardrails are treated as standardized, reusable patterns. These assets are stored in the TOGAF Architecture Repository to ensure consistency across different business domains.
• TOGAF provides the strategic roadmap, risk management (addressing issues like hallucinations), and architectural vision for AI.
• DAMA-DMBOK delivers operational rigor by ensuring high data quality for embeddings and establishing ethical governance and privacy standards.
• Architecture is shifting from traditional ETL toward Zero-ETL and data sharing. These paradigms favor logical integration—accessing data where it lives—to reduce costs, minimize latency, and increase agility.
• Advanced integration is managed through formal Data Contracts and Sharing Agreements. These machine-readable artifacts define quality rules, SLAs, and access terms, ensuring secure interoperability in decentralized environments.
• Using Architecture Boards and Data Councils, organizations monitor specific KPIs (like model accuracy and contract compliance). Feedback loops ensure that architectural patterns and principles are updated quarterly to reflect technological and regulatory changes.