Granted Patent B2

US 12,639,730 · App. 18/971,945 · Granted May 26, 2026

Inventor: Tero Heinonen (Sacramento, CA)

Assignee: LastBot Europe Oy

G06Q30/0271G06F16/2237G06F16/2425G06F16/2457G06F16/285G06N20/00G06Q30/015

Abstract

Claims (70)

1 . A method for generating data related to a recipient, the method comprising:

receiving, by at least one processor one or more input data sets relevant to the recipient;

identifying, by the at least one processor, one or more performance metrics associated with generating data related to the recipient, the performance metrics arranged to quantify one or more aspects of interest related to the recipient;

executing a generation process comprising:

analyzing, by a first artificial intelligence (AI) model executing on one or more processing units, at least one of the one or more input data sets in combination with one or more instructions, to generate first output data;

applying, by a machine learning model executing on the one or more processing units, the first output data to calculate one or more performance metrics, the machine learning model configured to optimize the calculated performance metrics; and

processing by a second AI model executing on the one or more processing units, the calculated performance metrics in combination with the one or more input data sets to generate second output data

iteratively repeating the generation process until a criterium for completion has been satisfied, wherein the criterium updates model parameters or embeddings to improve an accuracy of generated data; and

outputting, by the at least one processor, the second output data to a computer readable storage medium.

2 . The method according to claim 1 , wherein iteratively repeating the generation process comprises introducing the second output data directly as the first output data for further processing or as part of the input data sets, to be processed alongside other input data.

3 . The method according to claim 1

wherein at least one performance metric represents a predicted outcome in response to the first output data, the outcome encompassing actions, events, or changes in the physical or virtual world.

4 . The method according to claim 1 , wherein the first AI model and the second AI model are the same AI model or wherein the machine learning model, the first AI model, and the second AI model are the same AI model.

5 . The method according to claim 1 , wherein the one or more performance metrics are calculated using the first output data in conjunction with at least one of the one or more input data sets.

6 . The method according to claim 1 , comprising executing one or more of the steps in a plurality of AI contexts, and the executing comprises:

performing one or more of the steps in parallel across multiple AI contexts; or

performing one or more of the steps in parallel across multiple AI contexts,

selecting a subset of AI contexts based on the one or more performance metrics calculated for the plurality of AI contexts; or

performing one or more of the steps in parallel across multiple AI contexts,

selecting a subset of AI contexts based on the one or more performance metrics calculated for the plurality of AI contexts, and

discarding AI contexts with performance metrics not meeting the criterium for completion.

7 . The method according to claim 1 , wherein the first output data comprises a sequence of tokens generated by the first AI model, the tokens being selected to match a predefined pattern.

8 . The method according to claim 1 , wherein

the one or more performance metrics is provided to the AI model by encoding the one or more performance metrics as tokenized text; or

the one or more performance metrics is provided to the AI model as special tokens representing the values of the one or more performance metrics.

9 . The method according to claim 1 , wherein the method further comprises using the one or more performance metrics to select output data by:

maximizing an expected outcome based on the one or more performance metrics; or

implementing a reinforcement learning algorithm to balance exploration and exploitation strategies; or

aligning the selection with predefined user preferences or constraints.

10 . The method according to claim 9 , wherein the selection strategies comprise at least one of beam search, graph search, greedy search, contrastive decoding, hybrid search, A star, Q star, Q-learning, depth-first search, or breadth-first search.

11 . The method according to claim 1 , wherein the method further comprises one or more of the following:

conditioning at least one of the first AI model or the second AI model based on at least one of: the one or more performance metrics, at least one of the one or more input data sets, or at least one of the first output data or the second output data;

preprocessing the at least one of the one or more input data sets by performing at least one of: cleaning the input data; or enriching the input data;

encoding the at least one of the one or more input data sets as embeddings;

receiving one or more outcome data sets representing an outcome in response to the second output data and conditioning the machine learning model using at least one of the outcome data sets and at least one of: the one or more input data sets, the first output data, the second output data, the one or more performance metrics;

receiving one or more outcome data sets representing an outcome in response to the second output data and conditioning at least of the first AI model or the second AI model using at least one of the outcome data sets and at least one of: the one or more input data sets, the first output data, the second output data, the one or more performance metrics;

storing embeddings of the one or more input data sets and reusing unchanged embeddings in subsequent processing steps;

completion processing missing data in the one or more input data sets from existing data;

clustering embeddings and selecting or generating one or more representative embeddings of the cluster instead of the original embeddings in the cluster; or

changing the weight of data based on the time when the data was acquired or generated.

12 . A data processing system for generating data related to a recipient, the data processing system comprising a processor configured to perform the steps of the method of claim 1 .

13 . A machine-learning model for generating data related to a recipient, for use in the method of claim 1 , comprising:

a feature extraction layer configured to process a plurality of input data sets related to a recipient;

a prediction layer configured to calculate one or more prediction metrics associated with first output data generated by the first artificial intelligence (AI) model;

an optimization layer configured to refine predictions using the one or more prediction metrics; and

an output layer configured to generate second output data related to the recipient.

14 . A computer-implemented method of training the machine-learning model of claim 13 for generating data related to a recipient, wherein the method comprises:

receiving an input training dataset comprising a plurality of input data sets and corresponding target output data sets;

initializing the feature extraction layer to process input data sets and generate embeddings;

training the prediction layer to calculate prediction metrics associated with the target output data; and

optimizing the machine-learning model through iterative training cycles using loss minimization techniques to align predicted outputs with target outputs.

15 . A computer-implemented method of generating a training dataset for the machine-learning model of claim 13 , the method comprising:

receiving a plurality of input data sets related to a recipient;

generating target output data sets using a predefined set of rules or annotations;

associating input data sets with corresponding target output data sets; and

formatting the input and target data sets into a structured format compatible with the machine-learning model.

16 . A training dataset for use in the method of claim 14 of training the machine-learning model, comprising:

a plurality of input data sets related to a recipient, including structured and unstructured data; and

corresponding target output data sets that specify desired second output data related to the recipient.

17 . Use of the method of claim 1 for at least one of:

personalized content generation;

targeted marketing campaigns;

customer relationship management;

automated customer service;

automated customer representative;

automating contact center operations;

dynamic recommendation systems; and

predictive analytics for user engagement.

18 . A computer program for generating data related to a recipient, the computer program comprising instructions which, when executed by a processor, cause the processor to carry out the method of claim 1 .

19 . A computer program product for generating data related to a recipient, the computer program product comprising computer program code comprising instructions that, when executed by a processor, cause the processor to perform the method according to claim 1 .

Assignments (1)

Continuity (2)

References Cited (13)