Machine learning-based anomaly detection and action generation system for software development lifecycle management

Patent metadata

Field	Value
Publication number	US-12423657-B2
Application number	US-202519171161-A
Country	US
Kind code	B2
Filing date	Apr 4, 2025
Priority date	Mar 22, 2023
Publication date	Sep 23, 2025
Grant date	Sep 23, 2025

Abstract

Systems and methods for enhanced software development lifecycle management using a pipeline of machine learning and adaptive models for anomaly detection and termination. In some aspects, the system may generate a system data stream for an SDLC management platform that stitches together source data from multiple sources, generates a query to a first machine learning model for a user story identifier to generate trend information and anomalies, based on output from a second machine learning model including dynamic thresholds for the anomalies, determining at least one anomaly that satisfies a corresponding dynamic threshold, based on output from an adaptive model not including recommended actions to address the at least one anomaly, executing a kill switch for the corresponding user story identifier and generating for inclusion in the graphical user interface an indication of the kill switch.

First claim

We claim:

1. One or more non-transitory computer-readable media storing instructions, which when executed by at least one processor, perform operations comprising:
generating a system data stream for a software development lifecycle (SDLC) management platform that stitches together data from a plurality of computing systems;
receiving a first machine learning model trained based on past system data streams to generate trend information for one or more metrics and identify one or more anomalies in the trend information for an input user story identifier;
receiving a second machine learning model trained based on past trend information to generate one or more dynamic thresholds for the one or more anomalies in the trend information for the input user story identifier;
receiving an adaptive model trained on the past system data streams and the past trend information to generate one or more recommended actions to address the one or more anomalies in the trend information for the input user story identifier;
generating a query to the first machine learning model for each user story identifier of a plurality of user stories corresponding to the SDLC management platform;
based on output from the first machine learning model including (i) trend information for the one or more metrics and (ii) the one or more anomalies in the trend information for each user story identifier, generating a graphical user interface configured to be displayed at a client device, wherein the graphical user interface includes information for each user story of the plurality of user stories and corresponding progress tracking indicia;
based on output from the second machine learning model including the one or more dynamic thresholds for the one or more anomalies in the trend information for a corresponding user story identifier, determining an anomaly from the one or more anomalies that satisfies a dynamic threshold corresponding to the anomaly; and
based on output from an adaptive model not including one or more recommended actions to address the anomaly, executing a kill switch for the corresponding user story identifier and generating for inclusion in the graphical user interface the information for each user story of the plurality of user stories and corresponding progress tracking indicia and an indication of the kill switch being executed to address the anomaly.

2. A method for enhanced software development lifecycle (SDLC) management using a pipeline of machine learning and adaptive models for anomaly detection and termination, the method comprising:
receiving a system data stream for a software development lifecycle (SDLC) management platform that stitches together data from a plurality of computing systems;
generating a query to a first machine learning model for each user story identifier of a plurality of user stories corresponding to the SDLC management platform, wherein the first machine learning model generates trend information for one or more metrics and identifies one or more anomalies in the trend information for an input user story identifier;
based on output from the first machine learning model including (i) trend information for the one or more metrics and (ii) the one or more anomalies in the trend information for each user story identifier, generating one or more components of a graphical user interface configured to be displayed at a client device, wherein the one or more components of the graphical user interface comprise information for one or more user stories of the plurality of user stories and corresponding progress tracking indicia;
based on output from a second machine learning model including one or more dynamic thresholds for the one or more anomalies in the trend information for a corresponding user story identifier, determining an anomaly from the one or more anomalies that satisfies a dynamic threshold corresponding to the anomaly, wherein the second machine learning model generates the one or more dynamic thresholds for the one or more anomalies in the trend information for the input user story identifier; and
based on output from an adaptive model not including one or more recommended actions to address the anomaly, executing a kill switch for the corresponding user story identifier and generating for inclusion in the graphical user interface the information for each user story of the plurality of user stories and corresponding progress tracking indicia and an indication of the kill switch being executed to address the anomaly, wherein the adaptive model generates one or more recommended actions to address the one or more anomalies in the trend information for the input user story identifier.

3. The method of claim 2 , wherein the output from the adaptive model not including one or more recommended actions to address the anomaly includes:
processing, using the adaptive model, the anomaly in the trend information for the corresponding user story identifier, to generate one or more potential intervening actions and a confidence score for each potential intervening action, wherein the confidence score is based on a similarity of the anomaly to past anomalies and an effectiveness metric of past recommended actions to address the past anomalies; and
determining that none of the one or more potential intervening actions have a confidence score exceeding a threshold.

4. The method of claim 2 , wherein executing the kill switch for the corresponding user story identifier includes:
deactivating the corresponding user story identifier and preventing further progress until the anomaly is resolved.

5. The method of claim 2 , wherein executing the kill switch for the corresponding user story identifier includes:
executing a fallback mechanism to revert a user story for the corresponding user story identifier to a previous stable state.

6. The method of claim 2 , wherein the first machine learning model includes a recurrent neural network (RNN) trained based on past system data streams to capture temporal dependencies in the trend information.

7. The method of claim 2 , wherein the second machine learning model includes a gradient boosting machine (GBM) that uses feature importance scores to dynamically adjust thresholds for anomalies based on severity and frequency metrics for past anomalies.

8. The method of claim 2 , wherein the adaptive model includes a Large Language Model (LLM) trained on past system data streams and past trend information to generate the one or more recommended actions, wherein the LLM is integrated with a natural language processing (NLP) module to process user queries related to detected anomalies and recommended actions.

9. The method of claim 8 , wherein the LLM uses attention mechanisms to focus on a most relevant portion of the trend information when generating the one or more recommended actions to address the anomaly.

10. The method of claim 2 , wherein the adaptive model includes an enhanced Supervised Learning Model (SLM) trained on past system data streams and past trend information to generate the one or more recommended actions, wherein the enhanced SLM uses ensemble learning techniques to combine predictions of multiple models for improved accuracy.

11. The method of claim 10 , wherein the enhanced SLM uses a hierarchical model structure to handle different types of anomalies and generate context-specific recommended actions.

12. A system comprising at least one processor and one or more non-transitory, computer-readable media having instructions stored thereon that, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
receiving a system data stream for a software development lifecy

Assignees

• Citibank Na

Inventors

• Vinayaka Mayank
• Peterman Ryan
• Shanka Prasad Badari Narayana
• Lawton Richard
• Betgar Vitthal Ramling
• Danait Adar
• Kumar Balaji
• Kurian Robin J
• Sharma Maneet
• Veerasami Anantha

Classifications

• G06Q10/06

  Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling · CPC title

• G06Q10/101

  Collaborative creation, e.g. joint development of products or services · CPC title

• G06Q10/103 Primary

  Workflow collaboration or project management · CPC title

• G06Q10/063114

  Status monitoring or status determination for a person or group · CPC title