# US 2026/0087306 A1 · Deep learning system for dynamic prediction of order preparation times

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0087306 A1
- Document ID: US20260087306A1
- Status: Published application
- Kind: A1
- Application number: 19/406,017
- Filed: 2025-12-02
- Published: 2026-03-26
- Earliest claimed: 2021-04-30
- Assignees: Toast, Inc.
- Applicants: Toast, Inc.
- Inventors: Richard W. Jagolta; Adam Dziki; Seong Hyun Hwang
- Web page: https://patentgrep.com/patent/US20260087306A1

## CPC

- G06N 3/042: Knowledge-based neural networks; Logical representations of neural networks
- G06N 3/045: Combinations of networks
- G06N 3/0499: Feedforward networks
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G06N 3/096: Transfer learning
- G06Q 10/063: Operations research, analysis or management
- G06Q 10/06334: {using forecasting or predicting}
- G06Q 10/08: Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
- G06Q 20/202: {Interconnection or interaction of plural electronic cash registers [ECR] or to host computer, e.g. network details, transfer of information from host to ECR or from ECR to ECR}
- G06Q 20/203: {Inventory monitoring}
- G06Q 30/0633: {Managing shopping lists, e.g. compiling or processing purchase lists (shipping orders G06Q10/083; order filling G06Q10/087)}
- G06Q 50/12: Hotels or restaurants
- G07G 1/14: Systems including one or more distant stations co-operating with a central processing unit
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G07G: REGISTERING THE RECEIPT OF CASH, VALUABLES, OR TOKENS

## Abstract

A system and method are provided for dynamically updating a digital menu using deep-learning-based preparation-time prediction. A first deep learning neural network generates item-level embeddings for menu items based on historical preparation time records. Actual and estimated item-level preparation time vectors are generated using cosine-similarity weighting across subsets of the historical data. A second deep learning neural network is trained using the item-level vectors, ground-truth preparation times, and normalized non-categorical metadata processed through dense vector layers and concatenation. The trained network is executed to generate predicted item-level preparation times for menu items currently available for ordering. An updated digital menu including the predicted item-level preparation times is generated and transmitted to a client device, and the digital menu is automatically updated in real time on the client device in response to changes in the predicted preparation times.

## Claims

### Claim 1

A computer-implemented method for dynamically updating a digital menu using deep-learning-based preparation-time prediction, the method comprising:
retrieving, by one or more processors, a historical set of item-level preparation time records for menu items offered by a plurality of restaurants participating in a point-of-sale (POS) subscriber system;
training and executing, by the processors, a first deep learning neural network to generate item-level embeddings for each of the menu items;
for a first subset of the historical set, calculating actual item-level preparation time vectors based on corresponding historical item-level preparation time records;
for a second subset of the historical set, generating estimated item-level preparation time vectors based on historical item-level preparation time records for pluralities of menu items within the first subset, wherein each of the pluralities comprises highest-ranked item-level embeddings in the first subset that exhibit cosine similarities to corresponding item-level embeddings in the second subset, and wherein each estimated item-level preparation time vector comprises a weighted average of the highest-ranked item-level embeddings weighted by their cosine similarities;
retrieving a historical set of order-level preparation time records for preparation of orders from the POS subscriber system;
training a second deep learning neural network to predict item-level preparation times for the menu items, wherein inputs to the second deep learning neural network comprise (i) the item-level preparation time vectors, (ii) ground-truth actual preparation times for the orders, and (iii) metadata taken from the order-level preparation time records, wherein non-categorical metadata is normalized to a value between 0 and 1, passed through two fully-connected network layers to generate individual dense vector representations, the individual dense vector representations are concatenated to create concatenated dimensional features, and the concatenated dimensional features along with the item-level preparation time vectors and ground-truth actual preparation times are passed through three fully-connected network layers and one output layer to generate the item-level preparation times;
following training, executing the second deep learning neural network to generate predicted item-level preparation times for menu items currently available for ordering at a restaurant;
generating an updated digital menu comprising the menu items and the predicted item-level preparation times;
transmitting the updated digital menu to a client device associated with a guest; and
in response to detecting a change in at least one of the predicted item-level preparation times, updating at least a portion of the digital menu on the client device in real time without requiring manual refresh.

### Claim 2 (depends on 1)

The method of claim 1, wherein the first deep learning neural network comprises an enhanced Bidirectional Encoder Representations from Transformers (BERT) model.

### Claim 3 (depends on 1)

The method of claim 1, wherein the metadata comprises short-term kitchen load, total cost, dining option, or hour-of-day information.

### Claim 4 (depends on 1)

The method of claim 1, wherein the executing of the second deep learning neural network is performed every two seconds.

### Claim 5 (depends on 1)

The method of claim 1, wherein updating the digital menu comprises updating visual indicators of item-level preparation times on the client device.

### Claim 6 (depends on 1)

The method of claim 1, wherein the transmitting comprises transmitting incremental update messages including only the menu items for which the predicted item-level preparation times have changed.

### Claim 7 (depends on 1)

The method of claim 1, further comprising predicting an order-level preparation time for an order selected via the client device based on the predicted item-level preparation times, and displaying the predicted order-level preparation time on the digital menu.

### Claim 8

A non-transitory computer-readable storage medium storing instructions that, when executed by one or more processors, cause the processors to perform a method for dynamically updating a digital menu using deep-learning-based preparation-time prediction, the method comprising:
retrieving a historical set of item-level preparation time records for menu items offered by a plurality of restaurants participating in a POS subscriber system;
training and executing a first deep learning neural network to generate item-level embeddings for the menu items;
calculating actual item-level preparation time vectors for a first subset of the historical set;
generating estimated item-level preparation time vectors for a second subset of the historical set using cosine-similarity weighted averaging of highest-ranked item-level embeddings;
retrieving a historical set of order-level preparation time records;
training a second deep learning neural network to predict item-level preparation times, wherein non-categorical metadata is normalized, passed through two fully-connected layers to generate dense vectors, concatenated, and processed along with the item-level preparation time vectors and ground-truth actual preparation times through three fully-connected layers and one output layer;
executing the second deep learning neural network to generate predicted item-level preparation times for menu items currently available for ordering;
generating an updated digital menu comprising the menu items and the predicted item-level preparation times;
transmitting the updated digital menu to a client device; and
updating at least a portion of the digital menu on the client device in real time in response to detecting a change in at least one of the predicted item-level preparation times.

### Claim 9 (depends on 8)

The non-transitory computer-readable storage medium of claim 8, wherein the first deep learning neural network comprises an enhanced BERT model.

### Claim 10 (depends on 8)

The non-transitory computer-readable storage medium of claim 8, wherein the metadata comprises short-term kitchen load, total cost, dining option, hour of day, day of week, or date.

### Claim 11 (depends on 8)

The non-transitory computer-readable storage medium of claim 8, wherein the predicted item-level preparation times are updated every two seconds.

### Claim 12 (depends on 8)

The non-transitory computer-readable storage medium of claim 8, wherein updating the digital menu comprises modifying graphical indicators corresponding to the menu items.

### Claim 13 (depends on 8)

The non-transitory computer-readable storage medium of claim 8, wherein the digital menu is rendered within a mobile application executed on the client device.

### Claim 14 (depends on 8)

The non-transitory computer-readable storage medium of claim 8, wherein transmitting the updated digital menu comprises transmitting incremental update messages containing only changed predicted preparation times.

### Claim 15

A computer program product for dynamically updating a digital menu using deep-learning-based preparation-time prediction, the computer program product comprising:
a non-transitory computer-readable medium having program code stored thereon, the program code comprising:
program instructions to retrieve historical item-level preparation time records for menu items offered by restaurants participating in a POS subscriber system;
program instructions to train and execute a first deep learning neural network to generate item-level embeddings for the menu items;
program instructions to calculate actual item-level preparation time vectors for a first subset of the historical set;
program instructions to generate estimated item-level preparation time vectors for a second subset of the historical set using cosine-similarity weighted averaging of highest-ranked item-level embeddings;
program instructions to retrieve historical order-level preparation time records;
program instructions to train a second deep learning neural network to predict item-level preparation times, wherein non-categorical metadata is normalized, passed through two fully-connected layers to generate individual dense vector representations, concatenated with item-level preparation time vectors and ground-truth actual preparation times, and processed through three fully-connected layers and one output layer to generate the item-level preparation times;
program instructions to execute the second deep learning neural network to generate predicted item-level preparation times for menu items currently available for ordering;
program instructions to generate an updated digital menu comprising the menu items and the predicted item-level preparation times;
program instructions to transmit the updated digital menu to a client device; and
program instructions to update at least a portion of the digital menu in real time on the client device in response to detecting a change in at least one of the predicted item-level preparation times.

### Claim 16 (depends on 15)

The computer program product of claim 15, wherein the first deep learning neural network comprises an enhanced BERT model.

### Claim 17 (depends on 15)

The computer program product of claim 15, wherein the predicted item-level preparation times are updated every two seconds.

### Claim 18 (depends on 15)

The computer program product of claim 15, wherein the digital menu is displayed within a native mobile application on the client device.

### Claim 19 (depends on 15)

The computer program product of claim 15, wherein updating the digital menu comprises modifying icons, graphical overlays, or animated indicators corresponding to preparation times.

### Claim 20 (depends on 15)

The computer program product of claim 15, further comprising program instructions to predict an order-level preparation time based on the predicted item-level preparation times and to display the predicted order-level preparation time on the digital menu.
