PARTNERS / DATABRICKS
AtScale models your metrics and compiles them into SQL that runs on your lakehouse. Unity Catalog keeps its permissions, and nothing leaves your workspace.
THE CHALLENGE
On-time delivery gets rebuilt in every notebook that’s ever needed it. One version counts by carrier, another by customer, and a third drops shipments that never had a promised date. They’re all reading the same tables, and none of them agree.
WHAT YOU GET
You get one semantic model above the gold layer. AtScale compiles it into Databricks SQL, so notebooks, dashboards and AI agents all read the same metric.
SEE IT IN DATABRICKS
The question arrives in plain language. AtScale finds the one modeled metric that answers it and shows the tables, the columns, the numerator, the denominator and the grain. AtScale ACE writes the Databricks SQL, and your lakehouse runs it.
Someone asks
"What is on-time delivery by carrier this month?"
↓ THE METRIC RESOLVES TO TABLES AND COLUMNS
One definition, from the semantic model
on_time_delivery · a ratio, so it never sums across carriers
numerator: shipments delivered on or before the promised date
denominator: all shipments
excludes: shipments with no promised date, and cancelled orders
grain: carrier · catalog: prod.supply · owner: Supply Chain↓ ATSCALE ACE WRITES THE SQL. DATABRICKS RUNS IT
94.2%
September to date · on-time delivery · all carriers
Computed in Databricks SQL under Unity Catalog · 88K rows scanned instead of 4.7B · nothing copied out
WHAT IT DOES
Whether a shipment without a promised date counts and how a ratio rolls up across carriers are decisions the supply chain team made once. They aren’t in the tables, so every notebook that reads those tables reinvents them.
AtScale keeps those decisions in one semantic model and compiles each request into Databricks SQL. Unity Catalog keeps enforcing its own permissions, the query runs on your compute, and what comes back is a metric an analyst doesn’t have to rework.
HOW IT CONNECTS
01
It’s one connection to your SQL warehouse.
02
AtScale queries as the person asking, so catalog permissions apply without a second access system.
03
Definitions live in your repo as SML, AtScale’s open semantic modeling language, and get reviewed like code.
04
Aggregates are built and refreshed in your workspace from the questions people ask.
MORE WAYS TO USE IT
An agent reads the metric, not the medallion tables underneath it.
Python, SQL and BI all resolve to one definition.
SML lives in the repo, so a definition change is a pull request.
ACE builds them from real query patterns and retires them when they stop getting used.
Permissions don’t get decided anywhere else.
The medallion architecture gets a definition layer on top, with no new tables to manage.
PROOF
Case Studies
Case Studies
Case StudiesOn the BIRD-Interact benchmark, adding AtScale made an LLM twice as accurate.
FAQ
No. ACE (the AI Computation Engine) compiles each request into Databricks SQL that runs on your compute, and aggregate tables stay in your workspace.
Unity Catalog stays in charge. AtScale queries as the person asking, so catalog permissions apply to every answer and there’s no second permission system to maintain.
Yes. Agents call the semantic model through AtScale’s MCP server and get the same definition your BI tools use. That includes Databricks agents, through AtScale’s Databricks Marketplace listing, and Claude, ChatGPT and Google Gemini.
As code. Definitions live in your repo as SML, AtScale’s open semantic modeling language, so a change is a pull request that gets reviewed like any other code. AtScale works with GitHub, GitLab, Azure DevOps and Bitbucket.
No. ACE builds them in your workspace from real query patterns and retires them when they stop getting used.
The workshop takes thirty days from kickoff to results.