Conversational ordering agent with constrained menu grounding

What is claimed is:

1. 1.
   Independent · Method

   A computer-implemented method of taking an order at a point-of-sale system, the method comprising: receiving, from a customer, an order utterance in natural language; generating, by a language model, one or more candidate order items from the order utterance; matching each candidate order item against entries of a live menu data structure for a store; rejecting any candidate order item that does not match an entry of the live menu data structure; populating a modifier and a price for each matched order item from the matched entry and not from an output of the language model; and committing an order comprising the matched order items to the point-of-sale system.

2. 2.

   The method of claim 1, wherein the order utterance is received as speech and transcribed before the generating.

3. 3.

   The method of claim 1, further comprising, in response to the rejecting, prompting the customer to choose among entries of the live menu data structure that are closest to the rejected candidate order item.

4. 4.

   The method of claim 1, wherein the matching comprises comparing a match score with the confidence threshold.

5. 5.

   The method of claim 1, wherein the live menu data structure excludes entries marked unavailable by a kitchen display system.

6. 6.

   The method of claim 5, further comprising updating the live menu data structure when an entry is marked unavailable during the order.

7. 7.

   The method of claim 1, wherein the matching completes within a time that is acceptable to the customer.

8. 8.

   The method of claim 1, wherein the language model is constrained to output identifiers selected from the live menu data structure.

9. 9.
   Independent · System

   A point-of-sale system comprising: one or more processors; and memory storing instructions that, when executed by the one or more processors, cause the system to: receive an order utterance in natural language; generate, using a language model, candidate order items from the order utterance; match each candidate order item against entries of a live menu data structure; reject any candidate order item without a matching entry; populate a modifier and a price for each matched order item from the matched entry; and commit the matched order items as an order.

10. 10.

    The system of claim 9, further comprising a microphone at an order station, wherein the order utterance is captured by the microphone.

11. 11.

    The system of claim 9, wherein the instructions further cause the system to display the matched order items for confirmation before committing.

12. 12.

    The system of claim 11, wherein the matched order items are displayed at the order station.

13. 13.

    The system of claim 9, wherein the live menu data structure differs between two stores of a single merchant.

14. 14.

    The system of claim 9, wherein the language model is fine-tuned on transcripts of past orders at the store.

15. 15.

    The system of claim 9, wherein the instructions further cause the system to log each rejected candidate order item with the order utterance that produced it.

16. 16.
    Independent · Non-transitory computer-readable medium

    A non-transitory computer-readable medium storing instructions that, when executed by a processor of a point-of-sale system, cause the processor to perform operations comprising: generating, by a language model, candidate order items from a natural-language order; matching each candidate order item against a live menu data structure; refusing any candidate order item without a matching entry; and populating a price for each matched order item only from the live menu data structure.

17. 17.

    The medium of claim 16, wherein the operations further comprise populating modifiers for each matched order item from the live menu data structure.

18. 18.

    The medium of claim 16, wherein the natural-language order is received as text through a messaging channel.

19. 19.

    The medium of claim 16, wherein the operations further comprise suggesting an add-on item only when the add-on item is an entry of the live menu data structure.

20. 20.

    The medium of claim 19, wherein the add-on item is selected based on items in the order.