Ai system for finance and insurance products for real-time interaction and generation of textual and conversational responses to customers queries and method relating thereto

AI system for Finance & Insurance products for real-time interaction and generation of textual and conversational response to customer's queries, comprising:

• user interface (390) displaying icons (110, 120, 130, 140, 150, 160, 170, 180, 190, 195) representing different Finance & Insurance (F&I) products wherein a customer can access information relating to any of finance & insurance products by clicking on said corresponding representative icon on the display of said user interface, after entering deal ID number (410), wherein said user interface also displays the image (105) of the vehicle being purchased or purchased by the customer, which image is captured from the deal ID of the customer and wherein said user interface also enables a customer to apply for discount coupons won by the customer on certain F&I products, in the games played by the customer during interaction with the system and wherein further said user interface also displays answers to questions frequently asked by the customer regarding F&I products;
• a modified Retrieval-Augmented Generation (RAG) module (430) incorporating novel modules and fine-tuned for F&I products for contextualization and rephrasing query wherein said novel modules includes a Langchain's MongoDB chat history function and a custom short-term window memory mechanism wherein said Langchain's MongoDB chat history function enables to retrieve entire chat history for a specific customer from MongoDB database (431) which is indexed by the customer's unique deal ID (410), and wherein custom short-term window memory mechanism enables to generate a customized short-term sliding window (432) of recent chats by means of Pythons'list slicing technique wherein text of present customer's query (405) is reviewed in the context of recent chats contained in said sliding window and is contextualized and rephrased (433) by API call to the Large Language Model (LLM) using said short-term window memory, said user query and prompt engineering;
• a module (500) for sentiment analysis and product extraction wherein for sentiment analysis, a fine-tuned transformer model (440), which has been trained on finance and insurance dataset using HuggingFaces's “Trainer” API in Python, is used to analyze the text of the rephrased contextualized query (433) to determine whether the query reflects ‘positive’ or ‘negative’ sentiment of the customer, and:
• when the sentiment of the said query is positive, the rephrased query is passed through vector store (510) which identifies the closely-matching product;
• and
• when the sentiment of the contextualized and rephrased query (433) is negative, the system extracts the F&I product based on MiniLM Product Extraction model (530).
• a module (540) for generating personalized trivia relating to statistical information relevant to the F&I product extracted as above for negative sentiment and for the geographical location of the customer thereby ensuring state-specific and contextually relevant statistics so as to enable a customer to take a well-informed decision for buying or not buying a F&I product.
• a module (470) for AI avatar generation for F&I products, for providing text response as well as an oral response wherein oral response is spoken by an avatar mimicking a person's personality and replicating the person's voice,
• a ‘Contextual Response Index (CRI)’ module (445) which stores pre-generated responses and the associated videos path to previous queries for which the response was previously generated by the system thereby enabling the system to quickly reference past dialogues and responses and wherein the module also leverages advanced similarity search algorithms to enable similarity match searches of the current customer query with the queries already present in the ‘Contextual Response Index (CRI)’;
• and
• wherein for enhancing the retrieval process, the embedded queries are integrated into said ‘Contextual Response Index (CRI)’ using Facebook AI Similarity Search (FAISS) function which is optimized for efficient similarity search and clustering dense vectors enabling rapid identification of the closest matches between the presently embedded query and the previously stored vectors in ‘Contextual Response Index (CRI)’.