Innovative Corporate Query Handling with Domain-Specific RAG

Large Language Models (LLMs) have significantly impacted Natural Language Processing, but face limitations in domain-specific query response generation. To overcome these issues, this work proposes a novel method that combines LLMs with Retrieval Augmented Generation (RAG). We have designed a hybrid model to improve the accuracy of responses to domain-specific questions by integrating data from minutes of meetings from an external knowledge store. It involves preprocessing meeting data, developing a retrieval mechanism using similarity search, and implementing a generation model based on the LLaMA-2 13B architecture. We evaluated our model using BLEU and ROUGE scores, demonstrating improved performance in closed-domain question answering. The BLEU score of 0.56037 and ROUGE scores (ROUGE-1: 0.7211, ROUGE-2: 0.6217, ROUGE-L: 0.7211) indicate the model's proficiency in generating accurate responses while reducing hallucinations.