Categories: Data Architecture and Enterprise Architecture

Data Architecture (and Data Governance)

Definition: how data is structured, moved, protected, and kept trustworthy across systems — for AI, this includes the pipelines, indexes, and training/grounding datasets that feed every model.

What's involved:

1. A source of truth per dataset — one authoritative place updates land
2. Data quality — validation, reconciliation counts, completeness checks
3. Structured, queryable stores over flat logs — logs are a view, never the truth
4. Lineage — where data came from and what transformed it
5. Versioning and freshness — content hashes, re-indexing on change, as-of dates
6. Audit trail — every movement recorded (who/what/when)
7. Classification and retention — sensitivity tiers with matching handling
8. Pipelines instrumented end to end — every step observable

The opener: "Underneath this is a data-architecture question — source of truth, quality checks, lineage, versioning, and audit. AI output can only be as trustworthy as the data pipeline behind it, so let me start there…"

Cite it when: the question involves pipelines, datasets, document corpora, stale or wrong data, integrations moving data, or "how do we keep the AI's knowledge current."

Enterprise Architecture (and Integration)

Definition: keeping many systems coherent — shared standards, reference patterns, and deliberate integration — so new capabilities strengthen the whole instead of fragmenting it.

What's involved:

1. Reference architectures — the blessed pattern per problem class; divergence is justified explicitly
2. Standards and guardrails — auth, logging, deployment, data handling: same everywhere
3. AI enters through the enterprise front door — no shadow AI stack
4. Integration patterns — APIs, contracts, queues; loose coupling over point-to-point
5. ADRs — significant decisions written down: context, options, choice, consequences
6. Architecture review — designs challenged early, when change is cheap
7. Alignment with the Systems Architect — propose in writing, decide together, own the delegated space fully

The opener: "From an enterprise-architecture standpoint, the rule is that AI comes in through the front door — our reference architecture, our auth, our logging, our deployment standards — and I'd validate the design with the Systems Architect via an ADR before building. Within that frame…"

Cite it when: the question involves fitting AI into existing systems, standards, working with the architect, modernization, or "how would this coexist with what we have."

The pairing move

"Two structural layers: the data architecture — source of truth, quality, lineage, versioning — and the enterprise architecture it plugs into — reference patterns, shared standards, reviewed via ADR." Data feeds the system; enterprise architecture houses it. For the transcript-service scenario, this pairing is nearly the whole answer's skeleton.