Bank of America Corporation

Emad Noorizadeh

Ramakrishna R Yannam

Priyank R Shah

Ravisha Andar

Sushil Golani

Saahithi Chillara

Rajan Jigish Jhaveri

Ion Gerald McCusker

Hari Gopalkrishnan

Isaac Persing

Ashwini Patil

Jennifer Russell

Donatus Asumu

Yogesh Raghuvanshi

Viju Kothuvatiparambil

Christopher Keith Restorff

Castigliana Cimpian

Dana Patrice Morrow Branch

Prejish Thomas

Bharathiraja Krishnamoorthy

Pravin Kumar Sankari Bhagavathiappan

Chris Welles

Joshua Maki

Jie Zhang Nissel

Devanshu Mukherjee

Aaron Michael Hosford

Emmanuel Dibia

A method for simulating a historical call is provided. Methods may receive a communication between a human caller and an interactive voice response system. The communication may have occurred within a production environment. The communication may include utterances communicated by the human caller, responses presented to the human caller by the response system and/or an original outcome of the communication. Methods may receive original components used by the response system to form the responses. The original components may include prediction models and/or parameters. Methods may input the communication and the original components into a simulator. Methods may enable a user to swap out the original components for modified components. Methods may modify the original outcome based on the swap out of the original components for the modified components. Methods may present the modified outcome and the modified components.

Simulator of production scenarios in a non-production downstream environment

Systems, methods, and apparatus are provided for integrating AI-powered bot-generated responses with an agent interface during a live session with a customer. In response to a customer request, a live chat session may be initiated with an agent at first platform that includes an agent interface. A parallel session may be initiated at a second platform that includes an interactive response system and AI engine. An input from a customer may be displayed at the first platform and may also be received at the second platform. The second platform may derive intent from the input and generate an AI-based response. The response may be displayed in a window at the first platform. The agent may approve, reject, or modify the generated response. Following agent approval, the response may be inserted into the live customer session.

Integration of AI-powered conversational messaging with a live agent interaction

Methods and systems for transcribing communications are provided. Methods may include receiving a communication. Methods may include splitting the communication into a plurality of communication segments. Each communication segment may include two or more words. Methods may include transcribing each segment included in the plurality of communication segments, in parallel. The transcribing may include using a transformer neural network to transcribe each segment included in the plurality of communication segments. Methods may include generating a transcription from the transcribing. The transcription may be generated by combining the transcription of each of the communication segments into a combined transcription. Methods may include correcting the combined transcription.

Performance optimization for real-time large language speech to text systems

Apparatus, methods and systems for contextual prediction processing is provided. Methods may include receiving a conversation from an entity. The conversation may include current utterance, previous utterances and details. Methods may include using an action-topic ontology to build, using data retrieved from the current utterance, a conversation frame that corresponds to the current utterance. Methods may include merging the conversation frame with data, retrieved from the previous utterances and the details, to generate a target conversation frame. Methods may include validating the target conversation frame to prevent looping over historic data in the event that the current utterance fails to add relevant information. Methods may include generating an enhanced contextual utterance based on algorithms and the target conversation frame. The enhanced contextual utterance may be used to understand the current utterance in a context of the conversation. Methods may include returning the enhanced contextual utterance to the entity.

Selection system for contextual prediction processing versus classical prediction processing

Apparatus, methods and systems for an utterance parsing system is provided. The system may include a receiver, a tokenizer, a semantic role classifier, an action topic ontology and a semantic relationship builder. The receiver may receive an utterance. The tokenizer may tokenize the utterance into tokens. The semantic role classifier may determine which tokens are candidates for semantic role classification. The semantic role classifier may identify a semantic role for each candidate token. The semantic role classifier may determine that one semantic role is an action, and that one semantic role is a topic. The semantic role classifier may determine a class that corresponds to the topic-token, and a class that corresponds to the action-token. The action topic ontology may store the classes and one or more relationships between the classes. The semantic relationship builder may determine a vector defining a relationship between the action-token and the topic-token.

Action topic ontology

A system for determining intent in a voice signal receives a first voice signal that indicates to perform a task. The system sends a first response that comprises a hyperlink associated with a particular webpage used to perform the task. The system receives a second voice signal that indicates whether to access the hyperlink. The system determines intent of the second voice signal by comparing keywords of the second voice signal with keywords of the first response. The system activates the hyperlink in response to determining that the keywords of the second voice signal correspond to the keywords of the first response.

System and method for speech processing based on response content

Methods for generating test case scenarios in response to natural language requests may be provided. Methods may receive a natural language formatted request for a test case scenario. The natural language formatted request may include a plurality of parameters. The plurality of parameters may define a scope of the test case scenario. Methods may extract, utilizing natural language processing and artificial intelligence, the plurality of parameters from the request. Methods may query logging applications for log entries that are similar to the role of the test case scenario. Methods may construct, using artificial intelligence, operating on the log entries, a test case scenario in computer language. The test case scenario may correspond to the natural language formatted request. Methods may test the scope of the test case scenario by executing the test case scenario.

Test case scenario real-time generator

An interactive voice response system for use with a real-time complaint identification system and dashboard may be provided. The interactive voice response system may receive and respond to human callers in real-time. Simultaneously, the interactive voice response system may use a model to transcribe the audio call and identify complaints within the audio calls. Identified complaints may be resolved by one of a plurality of resolution executables. One of the resolution executables may include registering the identified complaint with a complaint register.

Call center voice system for use with a real-time complaint identification system

Systems are provided for building semantic frames. Systems may include building a semantic frame using a machine learning algorithm. The algorithm may identify: an index number of a token, a semantic role classifier assigned to the token, a corresponding correlation value and an index number of one or more related tokens. The algorithm may also create a semantic frame using the identified information. Systems may include building semantic frames for multiple tokens within an utterance. Systems may include building semantic frames for a plurality of tokens within a plurality of utterances. The plurality of utterances may be components of a conversation. Systems may also include summarizing the conversation using the semantic frames.

Semantic frame builder

A hybrid system for natural language processing is provided. The system may include a transceiver operable to receive a query from a user. The query may include a stock portion integrated with a set of real-time conditions specific to the user. The transceiver may transmit the query to a model and to a response application. The model may receive the query and communicate with a data store. Based on historical data stored at the data store, the model may separate the stock portion from the set of real-time conditions, formulate a response to the stock portion and insert placeholders into the response for responses to the real-time conditions portion. The response application may receive the query and the response. The response application may communicate with private data stores to formulate responsive elements for the placeholders. The response application may insert the responsive elements into the placeholders to complete the response.

Dynamic response engine

A system for dynamic semantic role classification, through an entity's natural language process (NLP) pipeline is provided. The system may include assigning semantic role classifiers to tokens included in utterances received from user nodes. The system may include using a machine learning algorithm to assign the semantic role classifiers. The machine learning algorithm may assign the semantic role classifiers based on a calculated correlation value. The machine learning algorithm may use training and testing data sets to dynamically update the semantic role classifiers.

Dynamic semantic role classification

Methods and apparatus for parallel intent processing at a network of chatbots may be provided. A query may be received at a user interface. The query may be transmitted to multiple chatbots at the same time. Each chatbot may identify an intent for the query. Each chatbot may also identify a percentage of accuracy that the chatbot considers that the intent corresponds to the query. The chatbots may present the intents to the user interface. The user interface may rank the intents. The user interface may display the ranked intents to the user. The user interface may receive an intent selection from the user. The user interface may direct the user to the chatbot that identified the selected intent.

Executing a network of chatbots using a parallel bot approach

A system for horizontal scaling of retraining machine learning models across operational domains is provided. The system may reduce computational overhead associated model retraining. The system may include an artificial intelligence ('AI') engine that determines target machine learning models that need to be retrained in response to changed training data. The AI engine may assign daemons to the target models. The daemons may gather retraining requirements such as source code and training data. The daemons may schedule the target models for retraining on a CPU or a GPU based model training system.

Machine learning model training system

A method for filtering a plurality of agent-customer interactions to determine whether one or more of a plurality of agent-customer interactions should be stored in a library of Artificial Intelligence (AI) files related to an interactive voice response system (IVR) is provided. The method may include receiving an identification of a plurality of IVR flashpoints, monitoring and/or reviewing the plurality of agent-customer interactions, and determining whether one of the plurality of agent-customer interactions meets a threshold number of the IVR flashpoints. For each of the plurality of agent-customer interactions that meets a threshold number of the IVR flashpoints, the method may further direct the IVR to convert the interaction into an IVR workflow and store the IVR workflow in the library of AI related to IVR.

Agent assist design—autoplay

Methods for building and training a network of chatbots are provided. Methods may perform data analysis on a first chatbot in the network. The first chatbot may include a first domain of intents. The data analysis may identify and rank the intents in the first domain. The first domain may represent answers to a first domain of user queries. Methods may select a predetermined number of highest-ranking intents from the first domain based on the data analysis. Methods may input the selected intents into a second chatbot in the network. The second chatbot may include a second domain of intents. The second domain may represent answers to a second domain of user queries. Methods may input a hook into the second chatbot. The hook may include a trigger word and may correspond to the first domain. When the hook is triggered, the second chatbot may invoke the first chatbot.

Growing community of inventors

Plano, TX, United States of America

Emad Noorizadeh