Posts tagged "metadata-management"
- What the agent reads
An AI agent's answer is only as good as what it reads, and every platform keeps its semantic layer for AI agents inside its own walls. Write the meaning first.
- Context for AI, context by AI: metadata needs a confidence score
Why AI-generated metadata needs a confidence score: business context for AI comes from people building the conceptual data model; a local LLM still guesses.
- Keep AI-generated metadata marked after you accept it
Accepted AI-generated metadata keeps its data provenance: AI badge, confidence score and rationale stay in the data catalog so data governance sees each guess.
- A Model You Can Diff: Metadata Version Control for Data Teams
Pipeline code gets code review and an audit trail; metadata doesn't. Five metadata version control practices bring naming conventions under data governance.
- Six steps to your integration layer: from bronze to trusted gold
Bronze shipped and gold still isn't trusted. Here are six steps to the integration layer your medallion architecture is missing.
- Integration Layer: Pattern, Not Project
Data warehouse automation makes the integration layer a pattern, not a project: a catalog rendered through a Jinja template under one naming convention.
- Top-down and bottom-up data modeling: cart before the horse
Top-down and bottom-up data modeling: join the conceptual data model and business glossary to source system discovery, so staging tables carry data governance.
- Keep the model stable, change the template
Keep the model stable, change the template: metadata-driven code generation turns one governed model into native Databricks code, with other platforms next.
- The Mapping Is the Deliverable
The source to target mapping that turns a source system into the business model is usually never written down. Record it with who decided, when, and why.
- AI drafts the tedious part. You keep the decision.
AI-assisted data modeling in DeltaVault: skills draft keys, joins, views and descriptions as proposals you read first. Nothing writes until you accept.
- Meaning Leaks at Every Handoff
Business definitions in one tool, the technical data model in another, lineage in a third: meaning leaks at every handoff. DeltaVault keeps them in one model.
- You already own the stack
You already own the lakehouse, storage, and orchestration. The missing layer is data warehouse automation: business meaning in, Data Vault and workflows out.
- Edit the template, watch the code change
Jinja template code generation you can see: open DeltaVault's template playground, edit the template, and watch the generated code update live.
- Manage data, not plumbing
Declare table metadata once; DeltaVault generates create-table, merge and workflow code for Databricks, Snowflake or Fabric. That is data warehouse automation.
- Your Database Is Not Your Business
AI can infer a semantic layer from your technical exhaust. It cannot define your business ontology. Why a system of record is not a system of meaning.
- AI does not fix bad data models
AI data modeling aimed at your source schema returns a faster, shinier source schema. Business meaning comes from your own material, and sources map into it.
- AI-native, not AI-added
What separates an AI-native data catalog from a bolted-on chat panel: governed writes, and AI skills you can open, edit, and extend.
- Meaning before machinery
AI made pipelines cheap to build. The scarce thing now is business context for AI: what your business means. DeltaVault captures it once and builds from it.