Select Star (acquired by Snowflake)

📉 Why does AI struggle with your data even when it’s “well documented”? Because documentation ≠ context. When a table says customer_id, AI sees a string. You see a relationship, a metric driver, and an entire business process. That’s where context engineering comes in. Instead of just storing metadata, you curate and serve it so AI tools know not just what the data is, but what it means. With context engineering, your metadata becomes: 🧠 Understandable to LLMs (not just humans) 🔍 Queryable and dynamic 🔗 Connected to lineage, usage, and real-world context Read the blog to learn how to make your data AI-ready (and your AI actually helpful): 👉 https://lnkd.in/gQVzSCgv #ContextEngineering #MetadataMatters #AIReady

If you asked five dashboards for “active customer,” would you get one answer or five? 👀 We’re joining Snowflake, dbt Labs, and other industry leaders on the Open Semantic Interchange (OSI) so semantics travel with you: BI to AI, tool to tool. With Select Star, turn trusted BI models into AI-ready semantics. Why it matters: one set of governed metrics, faster migrations, day-one explainability. If these sound familiar, a semantic layer will help: • The same KPI has multiple names or definitions across tools • SQL is copied between dashboards with subtle differences • There is no clear owner or approval for core metrics • Triage and audits are slow because lineage is unclear • AI assistants give inconsistent answers to the same question 📖 Read more: https://lnkd.in/gQSvC8qS #OSI #SemanticLayer #DataGovernance #AIForData

🤖📊 The hidden cost of AI replacing dashboards? Your semantic layer better be bulletproof. AI tools are great at generating answers. But they don’t know which metrics matter or how your business actually defines them. Without a solid semantic layer, copilots can serve up insights based on stale logic, inconsistent definitions, or the wrong source tables. A good semantic layer does more than just define metrics. It: 🧠 Embeds business context directly into your data stack 📏 Aligns AI-generated insights with trusted definitions 🔗 Connects metrics to documentation, lineage, and usage It’s how you turn fast answers into reliable ones and make AI analytics actually useful. Tobias Macey and Shinji Kim discuss on the Data Engineering Podcast why this matters more than ever: https://lnkd.in/gNvKDdKW

🤔 What’s more valuable than data itself? The metadata that makes it usable. We’ve all heard the saying that data is the new oil. But just like oil, raw data isn’t worth much on its own. It needs context to be useful. That’s what metadata provides. It helps you understand: ✨ What the data is 📍 Where it came from and where it’s going 📊 How it’s being used 👤 Who owns it ✅ Whether it can be trusted This context is becoming critical as data teams grow, dashboards multiply, and AI tools begin generating SQL and insights on their own. Metadata is quickly becoming the most important (and most underinvested) layer of the modern data stack. Without it, teams are flying blind. With it, you’re building an AI-ready environment. Shinji Kim joins Sean Falconer on the Software Engineering Daily podcast to dive deeper into why metadata matters now more than ever: https://lnkd.in/de_5qEwy

🧠 I was testing out Select Star's MCP server using Claude Desktop over the weekend. To answer a business question, Claude: - Got 634 assets from search - Filtered to 35 relevant ones - Selected 3 core tables - Gave me the answer in under 10 seconds That's a lot of tables to consider before finding 3 tables to use. This is because we have a lot of duplicated tables between Snowflake, BigQuery, and Redshift. I didn't specify the data source, so it needed to evaluate them all, look for the right tables, validate relationships, and data quality in order to find the right one. This is a common pattern in data discovery - lots of exploration, but the final answer usually involves just a few core tables. Still, Claude took only 10 seconds to come back with the right query because of the following: ✅ Popularity scores prioritized what data to look at ✅ Example queries in metadata helped validate relationships and similar use cases ✅ Lineage confirmed the correct tables in one step ✅ Tags like "Certified" surfaced trusted sources The truth is, schema metadata alone is not going to give you the right SQL. I'm learning that MCP isn’t just about RAG or better prompts. It’s also about encoding how data teams think through data questions. ❓What’s the relevant dataset also used by others? ❓How do these datasets connect? ❓Has someone asked this before? ❓Is the data up to date? Getting this right means faster, more accurate, and more explainable answers from your AI copilots. This kind of metadata context is what sets great AI apart. The good news is - we've already done the hard work at Select Star to build up the metadata, with context that AI likes. How's your data MCP experience been?

Financial services companies are under growing pressure to manage sensitive information responsibly. 🔍Discover how Trellance used Select Star to bring clarity and control to its data governance practices.