Published Paper Details:

Organisations create a lot of unstructured data, like policies, manuals, and reports. Traditional keyword-based search doesn't capture semantic meaning and context, which makes it hard to find what you're looking for. Even though Large Language Models (LLMs) can understand natural language, they often give wrong or made-up answers when used alone, especially when it comes to new or domain-specific information. This paper presents an Enterprise Retrieval-Augmented Generation (RAG) system that integrates semantic retrieval through dense vector embeddings with context-aware LLM generation. The system handles documents in many formats, uses FAISS to index them, and then uses a GPT-4 backend to give accurate answers that come from the right source. It has a dual-mode architecture for strict retrieval and better explanations, as well as a backup system for queries that aren't very relevant. Tests on a dataset of 500 documents show that it works well, with an 88% retrieval rate,improved response time (1.2s vs. 1.9s baseline), and support for multiple file formats, making it suitable for auditable, enterprise environments.