Blink answers best when your prompt says what you are looking for, why you need it, and in what form you want the answer.
This guide collects the prompting habits that get the most out of your DataGalaxy catalog.
What is Blink?
Blink is a conversational assistant connected to your DataGalaxy catalog through MCP tools. It helps users:
- Find datasets, reports, glossaries, dashboards, tables, KPIs, and processes
- Understand business definitions and relationships
- Explore lineage and hierarchy
- Discover experts, owners, stewards, and governance roles
- Retrieve metadata and collaboration information
- Navigate the catalog using natural language
Blink has some limitations and cannot:
- Operate with all custom attributes
- Navigate via the UI
- Deal with platform administration
How to think about prompting Blink
Treat Blink like a knowledgeable data colleague. Good prompts explain:
- What you are looking for
- Why you need it
- Any business context
- Expected output format
The more business context you provide, the better Blink can choose between keyword search, semantic search, metadata exploration, lineage retrieval, and governance discovery.
How to write a good prompt
Start with business context
Weak prompt: customer table
Better prompt: Find datasets related to customer churn analysis for telecom subscribers.
Why? It includes a business topic, gives semantic meaning, and helps Blink use semantic search effectively.
Use natural language, not technical syntax
Blink is optimized for conversational requests.
Avoid: table:customer AND owner:finance
Prefer: Show me finance-owned customer reporting datasets.
Be explicit about the object type
If you know what you need, say it — dataset, dashboard, KPI, glossary term, process, report, data source.
Example: Find dashboards related to sales forecasting. This reduces ambiguity and improves precision.
Include business vocabulary
Blink understands business semantics better when prompts contain department names, project names, business domains, use cases, metrics, and regulatory language.
Good examples: “GDPR-sensitive customer information,” “Supply chain forecasting KPIs,” “Finance reconciliation process,” “Marketing attribution dashboards.”
Ask for relationships and context
Blink can retrieve linked objects and hierarchy information.
- Find dependencies: What objects are linked to the Customer Master dataset?
- Explore hierarchy: Show the parent business domain of the Revenue KPI.
- Understand lineage: What reports depend on the Sales Fact table?
Ask follow-up questions
Blink performs best in conversational workflows. For example:
- Find datasets related to customer retention.
- Which of these are owned by the marketing team?
- Show linked dashboards for the top 3 datasets.
Specify the desired output
You can guide Blink toward concise or detailed responses.
- Concise: Give me a short summary of the Payment Processing dataset.
- Detailed: Provide complete metadata and linked objects for the Customer 360 dashboard.
- Governance-focused: Who are the stewards and owners of critical finance datasets?
Use semantic search for concepts
Semantic search works best for topics, concepts, business meaning, and intent-based discovery.
Strong semantic queries: “datasets about customer lifetime value,” “objects related to fraud detection,” “GDPR compliance reporting,” “employee onboarding process.”
Weak semantic queries: cust_tbl_v2, rpt_001.
Use precise names when known
If you already know the object name, include it directly.
Example: Show full details for dataset “Customer Revenue Snapshot.” This helps Blink retrieve exact object metadata faster.
Ask for tags and technologies
Blink can retrieve governance and technical classifications.
- Tags: What tags are associated with the Supplier dataset?
- Technologies: Which Snowflake objects are related to customer analytics?
- Data sources: Show datasets sourced from SAP.
Break complex requests into steps
Large requests are easier to solve incrementally.
Instead of: Find all finance datasets with lineage, owners, technologies, comments, and linked dashboards.
Prefer:
- “Find finance datasets.”
- “Show owners and stewards.”
- “Retrieve linked dashboards.”
- “Display technologies used.”
Ask Blink to compare or prioritize
- “What are the main differences between these two KPIs?”
- “Which datasets appear most relevant for churn analysis?”
Use collaboration features
Blink can retrieve comments and tasks.
- “Show comments on the Revenue dashboard.”
- “Are there open tasks related to Customer Master?”
- “Create a comment requesting data quality validation.”
Avoid overly vague requests
Too vague: Show me data.
Better: Show me datasets related to quarterly revenue reporting in finance.
Which prompt templates can you reuse?
| Purpose | Example |
|---|---|
| Dataset Discovery | Find datasets related to [business topic] used by [department/team]. |
| Governance Lookup | Who owns and maintains [object name]? |
| Lineage Exploration | What downstream reports depend on [dataset/table]? |
| Semantic Discovery | Find objects related to [business concept or use case]. |
| Expert Discovery | Who are the stewards for [business domain]? |
| Metadata Retrieval | Show full details and linked objects for [object name]. |
Best practices at a glance
| Best Practice | Why It Helps |
|---|---|
| Add business context | Improves semantic relevance |
| Use natural language | Matches Blink's search behavior |
| Specify object types | Reduces ambiguity |
| Ask step-by-step | Improves precision |
| Use follow-up questions | Enables conversational discovery |
| Include business vocabulary | Enhances semantic search |
| Request relationships | Unlocks lineage and hierarchy |
| Be explicit about intent | Produces more targeted responses |
What does an excellent prompt look like?
Why this works: clear business objective, strong domain context, multiple related retrieval goals, and an explicit expected outcome.
What to remember
To get the best results from Blink:
- Think in terms of business intent
- Be specific when possible
- Use conversational follow-ups
- Combine business and technical language
- Ask for relationships, ownership, and governance context
Blink is most powerful when used as a data discovery and governance copilot, not just a keyword search engine.