The answer your rep spends all day finding, drafted in forty seconds.
A customer sends a messy technical inquiry. GRADE extracts its hard constraints, screens your catalogue with deterministic SQL — never a guess — and drafts a reply with a reason and a source line for every product. Your rep approves every send.
A qualified answer takes hours the rep does not have.
Four steps, and the rep is always the last one.
Parse
Extract typed constraints from the inquiry — pH as two floats, a VOC ceiling as a number, PFAS-free as a boolean — each tied to the customer's own words.
Screen
Filter the catalogue with SQL. A hard limit is a WHERE clause, never a similarity score. Nothing that fails a stated constraint can survive.
Rank
Order the survivors. The model only ever sees products that already passed the deterministic screen — it cannot resurrect a failed one.
Draft
Write a reply, with a reason and a data-sheet quote per product. Your rep edits and approves. Nothing sends automatically.
The guarantees are in what it will not do.
If a product has no value for a constrained field, it is flagged unknown for a rep to check — never quietly counted as a pass.
Every draft waits for a human. GRADE qualifies and drafts; the rep decides and sends. There is no autonomous outbound.
Every extracted number carries a verbatim quote from the source document. If the quote is not in the sheet, the value is rejected.
We publish our methodology, not a logo wall.
Deterministic screening with no model judgment on hard constraints. Verbatim provenance or the value is rejected. An eval gate that permits zero constraint violations, independently re-verified. The screening invariants proven across more than a million generated cases. We also say plainly what we do not have yet.
See it on your own catalogue.
A short call, your real data sheets, one inquiry screened end to end.