Ask the way you would ask a colleague
Answers come back as tables and charts, drawn from the definitions your business already computes. A chart shows the same rows as its table and carries the same badge.
Ask your warehouse. Get the insight, and the proof behind it.
Molecular AI answers plain-English questions with figures, tables and charts, from the definitions your business already computes. Every figure carries a badge computed from what was proved, never decided by the model.
Four tables called revenue. Which one is the answer from?
Ask an AI about your business and it will find a table and give you a number. It will not tell you whether finance signs off on that table, whether the logic behind it still holds, or whether anyone ever checked.
A confident wrong number costs more than no number. So every figure here says how far to trust it.
A governed analyst for the warehouse you already have. Ask in plain English. See how far to trust each figure. Approve what it does not know yet.
See how far to trust it
Every figure is badged verified, qualified, unverified or unsourced. Hover to open its warrant, the signed record of what was proved: what ran against your engine, what passed, what was skipped, and when it expires.
Approve what it does not know yet
When no definition covers a question, it proposes one and marks the answer unverified. A person approves the proposal, it is proved, and the same question comes back verified.
It starts from what your business already knows
Before answering anything, it reads the meaning already spread across your stack and labels every piece by where it came from.
Your code
dbt projects and folders of SQL: the models, joins and logic behind the numbers your business already computes.
Query history
The queries your team actually runs, read from the engine, show which tables and definitions are really in use.
Your repositories
The READMEs, docs folders and data dictionaries beside the code that say what the data means.
What you upload
Handbooks, glossaries, runbooks and reports, as PDF, Word, Markdown or text. Where a document and the warehouse disagree, a person is asked.
What your team answers
The scan raises questions it cannot settle on its own. Each answer becomes part of the context.
Every piece of context is labelled measured, derived, supplied by a person, or inferred, so you can see how much was guessed.
Built for the people a wrong number lands on
Finance
Trace a figure in the board pack to its definition, and see whether its proof still stands.
Operations
Ask about stores, orders and products in plain English. No ticket, and no guessing which table is current.
Data teams
The definitions stay yours. Approve what is added, see what is proved, and know what has lapsed before anyone asks.
Leadership
Know which numbers in front of you were proved against the warehouse, and which were not.
Watch the warehouse today. Rebuild it next.
Monitor · available now
Know when the data under a number moves
Monitor checks your raw tables for row counts, volume, nulls, format changes and freshness, and lays each run out as a grid from green to red. A check that could not run says so instead of passing. Warrants lapse after 14 days, so a proof that has gone stale shows as stale.
Rebuild · a later phase
Then, a simpler warehouse underneath
In a later phase, Molecular will use what it has read to propose a cleaner warehouse with fewer layers and the same definitions, generate it, and run it against your engine before anyone sees it.
Connected to the warehouse you already have
Works with
-
-
-
-
-
-
BigQuery
-
-
Start with one question.
Connect your warehouse and ask. Or book a call and we will set it up with you.