System

A system comprising:

• a communication interface including a network interface controller coupled to a packet-switched network and configured to transmit and receive data packets;
• a memory storing instructions, a speech recognition model, a natural language processing model, a machine learning classification model trained on stored case records, a generative neural network model comprising a transformer architecture with a plurality of self-attention layers, and a speech synthesis model;
• and
• circuitry comprising one or more processors coupled to the memory and configured to execute the instructions to:
• receive, via the communication interface, audio data of a voice communication session from a terminal device and convert the audio data into text data using the speech recognition model;
• analyze the text data using the natural language processing model to detect keywords or phrases indicating a possibility of fraudulent communication and to extract feature information;
• compute a risk score using the machine learning classification model based on the feature information, generate an embedding vector from the text data, and execute a similarity search against the stored case records to update the risk score;
• construct a prompt data structure encoding the text data, the risk score, and conversation history, transmit the prompt data structure to the generative neural network model, and receive response text;
• transmit the response text and the risk score to the terminal device via the communication interface;
• sequentially update the risk score in response to additional text data and dynamically adjust the prompt data structure based on the updated risk score;
• and
• transmit warning information to the terminal device when the risk score exceeds a threshold.

A method performed by circuitry of a server coupled to a packet-switched network via a communication interface, the method comprising:

• receiving, via the communication interface, audio data of a voice communication session from a terminal device and converting the audio data into text data using a speech recognition model;
• analyzing the text data using a natural language processing model to detect keywords or phrases indicating a possibility of fraudulent communication;
• comparing the text data with stored case records and computing a risk score based on a result of the comparison;
• and
• generating, based on a result of the analysis, response text for probing the possibility of fraudulent communication, and transmitting the response text to the terminal device via the communication interface as a proposed response or as an automated agent response.