Morgan Stanley Services Group Inc.

Mehak Mehta

Shailesh Gavankar

Suryakant Brahmbhatt

Aratrika Sarkar

Mayank Jain

Cheryl Fernandes

Sanjit Vijay Mehta

Melissa Kagaju

Rahul Suresh

Ashish Kumar Singh

Vijay Kumar Butte

Abhijit Anant Naik

Sourabh Kumar Janghel

Satya Verma

Meet Singh

Aditya Ramesh

Ayyapparaj Radhakrishnan Ganesan

Sushil Ninawe

Jayati Tripathi

Lokin Makkena

A system and method may create and manage a code repository structure and related files for a ML system such as a NN by executing code to create a version control management platform template commit including a pipeline folder comprising deploy files; a configuration folder comprising key-value pairs describing variables used as input to the neural network; a notebooks folder; a steps folder including ML or NN code; a tests folder comprising test scripts; and a utils folder; and executing code to create a tag associated with a pull request and alter version control management platform data such that the tag is displayed in conjunction with data describing the branch associated with the pull request. Executing code to create a tag may include searching a commit history for a unique commit number associated with a pull request, and based on this search creating a tag.

System and method for machine learning development

A device, system and method for analyzing audio speech signals to detect fraudulent calls to a contact center comprising splitting an audio recording of a call in real-time into a foreground speech signal attributed to a main speaker and a background audio signal, extracting audio features from the foreground speech signal and background audio signal, inputting the extracted audio features into an ensemble model comprising multiple different machine learning models co-trained to cumulatively detect fraud, wherein the multiple different machine learning models include: a speaker audio model to detect audio speech anomalies, a speaker intent model to classify intent of the main speaker, a synthetic voice detection model to identify a non-human entity, and a prosody model to detect voice intonation of the main speaker. A prediction may be output, by the ensemble model, indicating whether the call is fraudulent.

Audio speech signal analysis for fraud detection

A device, system and method for analyzing audio speech signals to detect fraudulent calls to a contact center comprising splitting an audio recording of a call in real-time into a foreground speech signal attributed to a main speaker and a background audio signal, extracting audio features from the foreground speech signal and background audio signal, inputting the extracted audio features into an ensemble model comprising multiple different machine learning models co-trained to cumulatively detect fraud, wherein the multiple different machine learning models include: a speaker audio model to detect audio speech anomalies, a speaker intent model to classify intent of the main speaker, and a prosody model to detect voice intonation of the main speaker. A prediction may be output, by the ensemble model, indicating whether the call is fraudulent.

Systems and methods for translating natural language to SQL on a custom enterprise data warehouse powered by Generative AI. With an embodiment of the present invention, a natural language question may be converted to a meaningful and accurate database query, e.g., SQL query, relevant to tables existing in an enterprise data warehouse. An embodiment of the present invention is directed to a comprehensive approach of transforming a natural language query to a focused SQL query using domain specific data models across firmwide metadata systems and data systems. In response to a user query, an embodiment of the present invention performs metadata analysis, targeted data retrieval and then SQL generation. An embodiment of the present invention may apply data warehousing standards and guidelines followed in the enterprise and provide a plug-and-play type architecture and solution that is scalable to large warehousing and other systems.

Natural language to SQL on custom enterprise data warehouse powered by generative artificial intelligence

A computer-implemented method for extracting data from separate databases and loading the data into a central data store without introducing additional security vulnerabilities is disclosed. The method begins by receiving a configuration comprising one or more transformation functions to be applied to one or more data fields. Values of the one or more data fields are extracted from the databases for a plurality of records, the values being previously tagged with metadata indicating at least a datatype for the values. Based at least in part on the configuration and the metadata tagging, a function from the one or more transformation functions is selected for each value. A second database with a central data store is populated with transformed data. Requestors are not permitted to access to pre-transformed data in the first one or more databases while being allowed access to at least some post-transformed data in the central data store.

Information compartmentalizing data store

A system and method for creating accessibility of any website or application for people with sight, hearing or speech disabilities. The system and method can include receiving input of the website or the application to be accessed and an indicator as to specific disabilities a user, scoring the website or the application for its accessibility based on the specific disabilities of the user, and if the score is below a threshold, determining an alternative form for the input of the website or the application to accommodate the specific disabilities of the user.

Systems and methods for accessible websites and/or applications for people with disabilities

A computer-implemented method for identifying a use anomaly potentially exposing sensitive data is disclosed. The method comprises receiving data comprising logs of a communication involving a computing device, where the logs comprise distinct logs of at least three communication abstraction levels. At least three anomaly classifiers are operated for logs from each of the at least three communication abstraction levels. An ensemble model is used to identify an anomaly in the communication, by processing output from each of the at least three anomaly classifiers. The various logs from a moment in time when the anomaly occurred are collated, and a graphical user interface is generated for reviewing the identified anomaly and collated logs. A human reviewer is then alerted that an anomaly has been identified.

Multi-level log analysis to detect software use anomalies

Systems and methods for predicting and preventing system incidents such as outages or failures based on advanced log analytics are described. A processing center comprising an incident prediction server and log database may receive application server logs generated by an application server and historical incident data generated by an incident database server. The processing center may be configured to cluster a subset of application server logs and based on the subset of application server logs and the incident data, determine in real time or near real time the likelihood of occurrence of an incident such as a system outage or failure.

Growing community of inventors

Jersey City, NJ, United States of America

Mehak Mehta