Granted Patent B2

US 12,614,092 · App. 18/360,778 · Granted Apr 28, 2026

Inventors: Anagha Rumade (Mountainview, CA); Sadra Amiri Moghadam (Mission San Jose, CA); Abhik Banerjee (Milpitas, CA); Gupta Gundlapalli (Frisco, TX); Srinivasa Murthy Basavaraju (San Ramon, CA)

Assignee: JP Morgan Chase Bank, N.A.

G06N5/04G06F3/067G06F8/34G06F8/60G06F8/70G06F8/77G06F9/4411G06N20/00H04L63/1425

Abstract

Claims (48)

1 . A method for developing and deploying a machine learning pipeline, the method comprising:

collecting different types of data, by at least one processor, from one or more data sources, wherein the different types of data include batch data and real-time data;

continuously processing, by at least one processor, the different types of data to build a plurality of layers of a Lambda application, wherein each layer is configured to be deployed independently of the other layers;

building, by the at least one processor, a modular machine learning pipeline in a modifiable reference file, wherein the modular machine learning pipeline defines a framework comprising a no-code/low-code parameterized machine learning framework;

building and integrating, by the at least one processor, a CI/CD (Continuous Integration and Continuous Delivery/Continuous Deployment) pipeline to deploy a source code into the modular machine learning pipeline;

deploying, by the at least one processor, the modular machine learning pipeline as a single deployment machine learning pipeline for different use cases by revising the source code of the CI/CD pipeline, while the framework of the single deployment machine learning pipeline remains unchanged;

wherein the plurality of layers are different from each other, and one of the plurality of layers is configured as a prediction-generating layer to generate, based on the different types of data, prediction data that are imported, unidirectionally and independently without modifying the prediction data, to at least two different layers of the plurality of layers, and wherein the at least two different layers of the plurality of layers include a batch inference layer and a real-time inference layer configured to operate asynchronously with respect to the prediction-generating layer; and

wherein the CI/CD pipeline is configured to build and test the source code before deploying the source code directly into at least one of the plurality of layers based on the source code, the CI/CD pipeline is configured to be integrated independently into each of the respective plurality of layers, and the CI/CD pipeline is configured to continuously and automatically introduce the source code into any stage of the respective plurality of layers that the CI/CD pipeline determines is an appropriate layer to receive the source code based on new changes to the source code; and

wherein the CI/CD pipeline is configured to identify an error condition in the modular machine learning pipeline, ensure a continuous flow of new features by introducing the source code into any stage of the respective plurality of layers to correct the error condition in the modular machine learning pipeline.

2 . The method of claim 1 , wherein each component of the framework comprises modular components configured to be deployed independently.

3 . The method of claim 1 , wherein the prediction-generating layer is an inference layer.

4 . The method of claim 3 , wherein building the modular machine pipeline further comprises sequentially vetting and integrating a first layer of the plurality of layers into the modular machine learning pipeline before the next layer is vetted and integrated into the modular machine learning pipeline.

5 . The method of claim 1 , wherein building the modular machine pipeline further comprises building an end-to-end pipeline including a data pre-processing stage, a model training stage, a model evaluation stage, and a model deployment stage.

6 . The method of claim 5 , wherein the data pre-processing stage comprises:

collecting the different types of data as raw data; and

implementing feature engineering, feature extraction, and feature selection to select and transform the raw data to features;

wherein the features being configured to build one or more machine learning models.

7 . The method of claim 6 , further comprising storing the features in a feature store.

8 . The method of claim 6 , wherein the model training stage comprises receiving the features to train and test the one or more machine learning models.

9 . The method of claim 8 , wherein the model evaluation stage comprises using historical data to evaluate a performance of the one or more machine learning models and to identify a best performing model based on the evaluated performance.

10 . The method of claim 9 , wherein the model deployment stage comprises deploying the best performing model for operation within a production environment such that the best performing model is configured to make predictions based on production environment data obtained from the production environment while operating the best performing model within the production environment.

11 . The method of claim 10 , wherein the model deployment stage comprises a cloud based deployment to a cloud environment.

12 . The method of claim 11 , wherein the cloud environment includes a Virtual Private Cloud (VPC) and one or more containers; and

wherein the VPC is hosting the one or more containers, and the one or more containers are running one or more applications.

13 . The method of claim 1 , wherein the CI/CD pipeline is configured to create parameterized code.

14 . The method of claim 1 , further comprises monitoring, by the at least one processor, the single deployment machine learning pipeline for degradation caused by drift.

15 . The method of claim 1 , further comprising receiving via a user device an input from a user having no knowledge of software development and operations or machine learning operations, wherein the input being configured to build the no-code/low-code parameterized machine learning framework.

16 . A system for developing and deploying a machine learning pipeline, the system comprising:

a receiving module, that when executed by at least one processor, is configured to collect different types of data from one or more data sources, wherein the different types of data include batch data and real-time data;

a reference module, that when executed by the at least one processor, is configured to continuously process the different types of data to build a modular machine learning pipeline including a plurality of module layers of a Lambda application, wherein each layer is configured to be deployed independently of the other layers and the reference module is built by combining the plurality of module layers, independently, wherein the plurality of layers comprises a batch layer, a real-time layer and an inference layer;

a CI/CD pipeline module, that when executed by the at least one processor, is configured to deploy a source code into the modular machine learning pipeline, independently;

wherein the reference module, that when executed by the at least one processor, is configured to be deployed as a single deployment machine learning pipeline for different use cases by revising the source code of a CI/CD pipeline, while a framework of the single deployment machine learning pipeline remains unchanged;

wherein the plurality of layers are different from each other, and one of the plurality of layers is configured as a prediction-generating layer to generate, based on the different types of data without modifying the prediction data, prediction data that are imported, unidirectionally and independently, to at least two different layers of the plurality of layers, wherein the at least two different layers of the plurality of layers include the batch inference layer and the real-time inference layer configured to operate asynchronously with respect to the prediction-generating layer; and

wherein the CI/CD pipeline is configured to build and test the source code before deploying the source code directly into at least one of the plurality of layers based on the source code, the CI/CD pipeline is configured to be integrated independently into each of the respective plurality of layers, and the CI/CD pipeline is configured to continuously and automatically introduce the source code into any stage of the respective plurality of layers that the CI/CD pipeline determines is an appropriate layer to receive the source code based on new changes to the source code; and

wherein the CI/CD pipeline is configured to identify an error condition in the modular machine learning pipeline, ensure a continuous flow of new features by introducing the source code into any stage of the respective plurality of layers to correct the error condition in the modular machine learning pipeline.

17 . The system of claim 16 , wherein the framework is configured having a no-code/low-code parameterized machine learning framework; and

a CI/CD pipeline module, that when executed by at least one processor, is configured to deploy a source code into the single deployment machine learning pipeline.

18 . The system of claim 16 , wherein the reference module is configured to be built by sequentially vetting and integrating a first layer of the plurality of layers into the modular machine learning pipeline before the next layer is vetted and integrated into the modular machine learning pipeline.

19 . The system of claim 16 , wherein the reference module is configured to be deployed in a cloud environment in a cloud agnostic and technology agnostic manner.

20 . A tangible computer-readable medium having stored thereon, computer executable instructions that, if executed by a computing device, cause the computing device to perform a method for developing and deploying a machine learning pipeline, the method comprising:

collect different types of data, by at least one processor, from one or more data sources, wherein the different types of data include batch data and real-time data;

continuously process, by at least one processor, the different types of data to build a plurality of layers of a Lambda application, wherein each layer is configured to be deployed independently of the other layers;

build, by the at least one processor, a modular machine learning pipeline in a modifiable reference file, wherein the modular machine learning pipeline defines a framework comprising a no-code/low-code parameterized machine learning framework;

build and integrate, by the at least one processor, a CI/CD pipeline to deploy a source code into the modular machine learning pipeline;

deploy, by the at least one processor, the modular machine learning pipeline as a single deployment machine learning pipeline for different use cases by revising the source code of the CI/CD pipeline, while the framework of the single deployment machine learning pipeline remains unchanged;

wherein the plurality of layers are different from each other, and one of the plurality of layers is configured as a prediction-generating layer to generate, based on the different types of data, prediction data that are imported, unidirectionally and independently without modifying the prediction data, to at least two different layers of the plurality of layers, and wherein the at least two different layers of the plurality of layers include a batch inference layer and a real-time inference layer configured to operate asynchronously with respect to the prediction-generating layer; and

wherein the CI/CD pipeline is configured to build and test the source code before deploying the source code directly into at least one of the plurality of layers based on the source code, the CI/CD pipeline is configured to be integrated independently into each of the respective plurality of layers, and the CI/CD pipeline is configured to continuously and automatically introduce the source code into any stage of the respective plurality of layers that the CI/CD pipeline determines is an appropriate layer to receive the source code based on new changes to the source code; and

wherein the CI/CD pipeline is configured to identify an error condition in the modular machine learning pipeline, ensure a continuous flow of new features by introducing the source code into any stage of the respective plurality of layers to correct the error condition in the modular machine learning pipeline.

Assignments (1)

Continuity (1)

References Cited (9)