Automated ordering system

1. A method comprising:

receiving, by a software agent executing on a server, a communication comprising a first utterance from a customer;

predicting, using an intent classifier, a first intent of the first utterance;

based on determining that the first intent is order-related, predicting, using a dish classifier, a cart delta vector based at least in part on the first utterance;

modifying a cart associated with the customer based on the cart delta vector;

predicting, using a dialog model, a first dialog response based at least in part on the first utterance; and

providing the first dialog response, by the software agent, to the customer using a text-to-speech converter.

2. The method of claim 1 , further comprising:

receiving, by the software agent, a second utterance from the customer;

predicting, using the intent classifier, a second intent of the second utterance;

based on determining that the second intent is menu-related, retrieving menu-related information based at least in part on the second utterance;

predicting, using the dialog model, a second dialog response based at least in part on the second utterance and the menu-related information; and

providing the second dialog response to the customer using the text-to-speech converter.

3. The method of claim 1 , wherein modifying the cart associated with the customer based on the cart delta vector comprises:

adding a new item to the cart;

deleting a current item from the cart;

modifying an existing item in the cart; or

any combination thereof.

4. The method of claim 1 , wherein predicting, using the dialog model, the first dialog response based at least in part on the first utterance comprises:

predicting the first dialog response based on a plurality of candidate responses and based on a dialog policy and an order context.

5. The method of claim 4 , wherein the order context comprises:

an interaction history between the customer and the software agent;

a cart state of the cart associated with the customer; and

a conversation state of a conversation between the customer and the software agent,

the conversation including the first utterance and the first dialog response.

6. The method of claim 1 , further comprising:

receiving, by the software agent, a third utterance from the customer;

predicting, using the intent classifier of the software agent, a third intent of the third utterance;

based on determining that the third intent is order-related, closing the cart;

receiving payment information from the customer; and

initiating order fulfillment of items in the cart.

7. The method of claim 1 , wherein receiving, by the software agent executing on the server, the communication comprising the first utterance from the customer comprises:

receiving audio data that includes the first utterance;

converting the audio data to text using a speech-to-text converter; and

performing post processing on the text to create a corrected utterance.

8. A server comprising:

one or more processors; and

one or more non-transitory computer readable media storing instructions executable by the one or more processors to perform operations comprising:

receiving, by a software agent executing on the server, a communication comprising a first utterance from a customer;

predicting, using an intent classifier, a first intent of the first utterance;

based on determining that the first intent is order-related, predicting, using a dish classifier, a cart delta vector based at least in part on the first utterance;

modifying a cart associated with the customer based on the cart delta vector;

predicting, using a dialog model, a first dialog response based at least in part on the first utterance; and

providing the first dialog response, by the software agent, to the customer using a text-to-speech converter.

9. The server of claim 8 , wherein the operations further comprise:

receiving, by the software agent, a second utterance from the customer;

predicting, using the intent classifier, a second intent of the second utterance;

based on determining that the second intent is menu-related, retrieving menu-related information based at least in part on the second utterance;

predicting, using the dialog model, a second dialog response based at least in part on the second utterance and the menu-related information; and

providing the second dialog response to the customer using the text-to-speech converter.

10. The server of claim 8 , wherein modifying the cart associated with the customer based on the cart delta vector comprises:

adding a new item to the cart;

deleting a current item from the cart;

modifying an existing item in the cart; or

any combination thereof.

11. The server of claim 10 , wherein:

the new item, the current item, and the existing item correspond to menu items in a menu associated with a restaurant.

12. The server of claim 8 , wherein predicting, using the dialog model, the first dialog response based at least in part on the first utterance comprises:

predicting the first dialog response from a plurality of candidate responses based on a dialog policy and an order context.

13. The server of claim 12 , wherein the order context comprises:

an interaction history between the customer and the software agent;

a cart state of the cart associated with the customer; and

a conversation state of a conversation between the customer and the software agent, the conversation including the first utterance and the first dialog response.

14. The server of claim 8 , wherein the operations further comprise:

receiving, by the software agent, a third utterance from the customer;

predicting, using the intent classifier of the software agent, a third intent of the third utterance;

based on determining that the third intent is order-related, closing the cart;

receiving payment information from the customer; and

initiating order fulfillment of items in the cart.

15. A memory device to store instructions executable by one or more processors to perform operations comprising:

receiving, by a software agent executing on a server, a communication comprising a first utterance from a customer;

predicting, using an intent classifier, a first intent of the first utterance;

based on determining that the first intent is order-related, predicting, using a dish classifier, a cart delta vector based at least in part on the first utterance;

modifying a cart associated with the customer based on the cart delta vector;

predicting, using a dialog model, a first dialog response based at least in part on the first utterance; and

providing the first dialog response, by the software agent, to the customer using a text-to-speech converter.

16. The memory device of claim 15 , wherein the operations further comprise:

receiving, by the software agent, a second utterance from the customer;

predicting, using the intent classifier, a second intent of the second utterance;

based on determining that the second intent is menu-related, retrieving menu-related information based at least in part on the second utterance;

predicting, using the dialog model, a second dialog response based at least in part on the second utterance and the menu-related information; and

providing the second dialog response to the customer using the text-to-speech converter.

17. The memory device of claim 15 , wherein modifying the cart associated with the customer based on the cart delta vector comprises:

adding a new item to the cart;

deleting a current item from the cart;

modifying an existing item in the cart; or

any combination thereof.

18. The memory device of claim 15 wherein predicting, using the dialog model, the first dialog response based at least in part on the first utterance comprises:

predicting the first dialog response based on a plurality of candidate responses and based on a dialog policy and an order context.

19. The memory device of claim 18 , wherein the order context comprises:

an interaction history between the customer and the software agent;

a cart state of the cart associated with the customer; and

a conversation state of a conversation between the customer and the software agent, the conversation including the first utterance and the first dialog response.

20. The memory device of claim 15 , wherein the operations further comprise:

receiving, by the software agent, a third utterance from the customer;

predicting, using the intent classifier of the software agent, a third intent of the third utterance;

based on determining that the third intent is order-related, closing the cart;

receiving payment information from the customer; and

initiating order fulfillment of items in the cart.