System and method for generating a chat response using a large language model

1 . A method for generating a large language model (LLM)-based chatbot response to an input question, comprising:

generating, using a first specific-trained language model (STLM), a set of filters based on an input question that relates to at least one sales call, wherein the generated set of filters refers to at least one metadata in a call dataset;

filtering the call dataset by applying the set of filters to the metadata included in the call dataset, wherein the call dataset further includes transcripts and their respective metadata;

embedding the input question into a vector representation;

comparing the vector representation of the input question to vector representations of textual information stored in the call dataset to retrieve a vector representation of a target dataset, wherein the target dataset contains an answer to the input question;

engineering a prompt to provide a single coherent command including information in the target dataset and the input question; and

feeding the engineered prompt to a second STLM to output the answer to the input question.

2 . The method of claim 1 , further comprising:

retrieving a contextual information from at least one data source; and

enriching the input question with the retrieved contextual information, and wherein the prompt is further engineered based on retrieved contextual information.

3 . The method of claim 2 , wherein the at least one data source is a CRM system.

4 . The method of claim 1 , wherein embedding the input question into a vector representation further comprises:

applying an embedding model on the input question, wherein the embedding model implements one or more sentence-embedding models.

5 . The method of claim 1 , wherein embedding the input question into a vector representation further comprises:

applying a single vector embedding model on the input question, wherein a single vector embedding model is based on two distinct vector embedding models.

6 . The method of claim 1 , wherein embedding the input question into a vector representation further comprises:

applying a large language model (LLM) to the input question.

7 . The method of claim 1 , further comprising:

rephrasing the input question using a third STLM.

8 . The method of claim 7 , wherein the first STLM, the second STLM, and the third STLM are same LLM trained on same sales data.

9 . The method of claim 1 , wherein an answer space of the input question is the call dataset and wherein the sales call further includes sales meetings and sales transactions.

10 . A non-transitory computer-readable medium having stored thereon instructions for causing a processing circuitry to execute a process, the process comprising:

generating, using a first specific-trained language model (STLM), a set of filters based on an input question that relates to at least one sales call, wherein the generated set of filters refers to at least one metadata in a call dataset;

filtering the call dataset by applying the set of filters to the metadata included in the call dataset, wherein the call dataset further includes transcripts and their respective metadata;

embedding the input question into a vector representation;

comparing the vector representation of the input question to vector representations of textual information stored in the call dataset to retrieve a vector representation of a target dataset, wherein the target dataset contains an answer to the input question;

engineering a prompt to provide a single coherent command including information in the target dataset and the input question; and

feeding the engineered prompt to a second STLM to output the answer to the input question.

11 . A system for generating a large language model (LLM)-based chatbot response to an input question, comprising:

a processing circuitry; and

a memory, the memory containing instructions that when executed by the processing circuitry, configure the system to:

generate, using a first specific-trained language model (STLM) a set of filters based on an input question that relates to at least one sales call, wherein the generated set of filters refers to at least one metadata in a call dataset;

filter the call dataset by applying the set of filters to the metadata included in the call dataset, wherein the call dataset further includes transcripts and their respective metadata;

embed the input question into a vector representation;

compare the vector representation of the input question to vector representations of textual information stored in the call dataset to retrieve a vector representation of a target dataset, wherein the target dataset contains an answer to the input question;

engineer a prompt to provide a single coherent command including information in the target dataset and the input question; and

feed the engineered prompt to a second STLM to output the answer to the input question.

12 . The system of claim 11 , wherein the system is further configured to:

retrieve a contextual information from at least one data source; and

enrich the input question with the retrieved contextual information, and wherein the prompt is further engineered based on retrieved contextual information.

13 . The system of claim 12 , wherein the at least one data source is a CRM system.

14 . The system of claim 11 , wherein the system is further configured to:

apply an embedding model on the input question, wherein the embedding model implements one or more sentence-embedding models.

15 . The system of claim 11 , wherein the system is further configured to:

apply a single vector embedding model on the input question, wherein a single vector embedding model is based on two distinct vector embedding models.

16 . The system of claim 11 , wherein the system is further configured to:

apply a large language model (LLM) to the input question.

17 . The system of claim 11 , wherein the system is further configured to:

rephrase the input question using a third STLM.

18 . The system of claim 17 , wherein the first STLM, the second STLM, and the third STLM are same LLM trained on same sales data.

19 . The system of claim 11 , wherein an answer space of the input question is the call dataset and wherein the sales call further includes sales meetings and sales transactions.