Intuit, Inc.

Noa Haas

Oren Sar Shalom

Alexander Zhicharevich

Rami Cohen

Shimon Shahar

Adi Shalev

Liron Hayman

Noah Eyal Altman

Tzvika Barenholz

Yair Horesh

Yehezkel Shraga Resheff

Yaron Sheffer

Gleb Keselman

Natalie Bar Eliyahu

Shir Meir Lador

Talia Tron

Shon Mendelson

Alexander Zicharevich

Nitzan Gado

Oren Dar

Balachandra Maddina

Asaf Brill

Techniques described herein involve dynamic prompt engineering using machine learning models. Embodiments include generating, using a machine learning model, a first output based on a sample data set. Embodiments include generating, using a fine-tuned machine learning model, a second output based on the sample data set, wherein the fine-tuned machine learning model is trained by iteratively adjusting parameters of the model based on comparing training inputs passed through multiple layers of the model to training outputs from the model. Embodiments include comparing the first output to the second output, wherein the comparing comprises identifying discrepancies between the first and the second output. Embodiments include automatically generating natural language instructions for providing to the machine learning model based on the comparing. Embodiments include performing an action using the machine learning model based on the natural language instructions and/or sending the natural language instructions to system components via a network communication protocol.

Machine learning model dynamic prompt engineering

A method classifies feedback from transcripts. The method includes receiving an utterance from a transcript from a communication session and processing the utterance with a classifier model to identify a topic label for the utterance. The classifier model is trained to identify topic labels for training utterances. The topic labels correspond to topics of clusters of the training utterances. The training utterances are selected using attention values for the training utterances and clustered using encoder values for the utterances. The method further includes routing the communication session using the topic label for the utterance.

Classifying feedback from transcripts

A computing system identifies an incoming voice call from a user device to an agent device associated with the computing system. The computing system generates a transcription of the incoming voice call using one or more natural language processing techniques. The computing system extracts a problem description from the transcription. The problem description indicates a topic for the incoming voice call. A first machine learning model estimates a situation vector from the problem description. A second machine learning model identifies a pre-existing situation vector that closely matches the estimated situation vector. The computing system retrieves a situation description that corresponds to the identified pre-existing situation vector.

Methods and systems for generating problem description

Systems and methods for automatically identifying problem-relevant sentences in a transcript are disclosed. In an example method, a transcript may be received of a first support call. A region of the first support call transcript may be identified, and first customer utterances may be detected in the region using a trained classification model. A trained regression model may estimate a relevancy to the problem statement of each of the first customer utterances, and one or more most problem-relevant statements may be selected from the first customer utterances, based on the estimated relevancies.

Extracting customer problem description from call transcripts

Operating an encoder with double decoder machine learning models include executing, on a transcript, an encoder machine learning model to generate an encoder output, and executing a situation decoder machine learning model on the encoder output to obtain a situation model output having a situation identifier, and executing a trouble decoder machine learning model using the encoder output to obtain a trouble identifier. The method further includes outputting the situation identifier and the trouble identifier.

Encoder with double decoder machine learning models

A method and system identify characteristics of financial transaction description strings. The method and system trains an analysis model with a machine learning process to classify financial transaction description strings. The analysis model generates a table that indicates the portions of the financial transaction description strings that were relevant in classifying the financial transaction description strings and the portions that were not relevant.

Method and system for identifying characteristics of transaction strings with an attention based recurrent neural network

A method that involves receiving a set of first log records. The set of first log records are duplicated using key operation pairs to generate a set of second log records from the set of first log records. The set of second log records are duplicated using a second set of intervals to generate a set of third log records from the set of second log records. The set of third log records are aggregated using the second set of intervals to generate a set of aggregated log records. The set of aggregated log records includes an aggregated log record comprising a number indicating a number of events from the set of first log records that have the key, that have the operation, and that occurred during an interval of the second set of intervals. The operation is identified by comparing the aggregated log record to a server log record.

Systems and methods for log aggregation

A method for detecting an unauthorized activity on a computer system involves obtaining current time stamps for a first type of access event related to the computer system, determining a current count of the first type of access event using the current time stamps, and predicting an expected count of the first type of access event using a current count of time stamps and a predictive model. The method further involves obtaining an actual count of the first type of access event, executing a first comparison of the actual count with the expected count, determining, based on a test comprising the first comparison, that the unauthorized access to the computer system occurred, and issuing an alert indicating the unauthorized activity occurred.

Growing community of inventors

Tel Aviv, Israel

Noa Haas