Watch Grok work out where a solar panel business can improve efficiency, using AtScale’s semantic layer as the only path to the data.
Enterprise AI gets data wrong when it has to guess. A model pointed at a raw warehouse has to work out how thousands of tables join and which metric definition is the right one, and then it has to write that SQL correctly every time. On the public BIRD-Interact benchmark, that approach gets about one question in five right.
In this demo, Grok connects to AtScale over MCP and asks one open question. It queries what looks like a single table, and AtScale turns that into the full SQL against PostgreSQL, using pre-built aggregates where they fit. Every query came back in under five milliseconds. Each answer also ties back to metrics certified in the semantic layer, so you’ll get the same number on Thursday that you got on Monday.
What You’ll Learn
- How Grok connects to AtScale’s semantic layer through the MCP server
- How AtScale turns a simple business query into the full SQL your platform needs
- How aggregate substitution kept every query in the demo under five milliseconds
- How AtScale asks a clarifying question when a request is ambiguous
- How certified metrics keep AI answers consistent from one day to the next
