Clarity: An Intelligent Enterprise Assistant Leveraging Retrieval-Augmented Generation

Abstract

Enterprise knowledge management is a nightmare. This information becomes divided, the search engines are rubbish and everybody loses time. Research indicates that knowledge workers take up to 36 percent of their time to hunt down the internal documents and that firms continue to drown in a pool of disjointed applications, which further complicates the situation. In this paper, we present CLARITY, a new AI involving assistant, which employs Retrieval-Augmented Generation (RAG) and chill semantic searching technology and live analytics. The system is a combination of dense vector search and keyword matching and achieves 94.2% accuracy compared to 86.5% of the older RAG systems. CLARITY puts vectors into ChromaDB, does smarter understanding with BGE-large-en-v1.5 embeddings, and answer generation with Llama-3.1-70B, which is aware of all the context. The result on 10,000 queries is an F1-score of 92.1-percent, an average response time of 2.1-seconds, and a reduction in errors by 57-percent. Security is good: role-based access, 2-factor authentication and activity logs. The main lessons: an RAG hybrid design that is efficient in companies, a real-time dashboard to gain insights, and a simple scaling. The comparison reveals that CLARITY outperforms the current solutions by 15% in accuracy, 13.2% in retrieval accuracy and reduces operational costs by 81% over API based solutions.