AI BASED DOCUMENT QUERY SYSTEM

  This paper presented a domain-agnostic, multimodal, and explainable Retrieval-Augmented Generation (RAG) framework designed to operate across heterogeneous enterprise document repositories. The proposed system integrates multimodal document ingestion, adaptive preprocessing, hybrid sparse–dense retrieval, reranking, and context-grounded answer generation within a unified and scalable architecture. By combining semantic vector search with keyword-based indexing, the framework improves retrieval robustness across both concept-driven and exact-match queries. The incorporation of provenance tracking and evidence-grounded generation enhances transparency and reduces hallucination, thereby increasing trustworthiness in enterprise applications.   The implementation through a functional dashboard demonstrates practical deploy ability and real-time incremental indexing capability. Overall, the proposed system provides a reliable and scalable solution for enterprise knowledge access, enabling accurate, explainable, and source-traceable information retrieval in dynamic document environments.

[1]      Kim, S., Song, H., Seo, H. and Kim, H., 2025. Optimizing retrieval strategies for financial question answering documents in retrieval-augmented generation systems. arXiv preprint arXiv:2503.15191.. [2]   Choi, C. et al. (2025). FinDER: Financial dataset for QA and RAG evaluation. arXiv:2504.15800. [3]   Wang, X. et al. (2025). FinSage: A multi-aspect RAG system for financial filings QA. arXiv:2504.14493. [4]    Setty, S. et al. (2024). Improving retrieval for RAG-based QA models on financial documents. arXiv:2404.07221.  [5]     Song, R., Li, Y., Giunchiglia, F. and Xu, H., 2026. Retrieval augmentation for out-of-distribution robustness in non-knowledge intensive in-context learning. Science China Information Sciences, 69(2), p.122108. [6]   Qiang, M., Wang, Z., Li, S. and Zhou, G., 2026. Exploring Knowledge Filtering for Retrieval-Augmented Question Answering. IEEE Transactions on Audio, Speech and Language Processing. [7]       Eponon, A.A., Shahiki-Tash, M., Ramos, L., Maldonado-Sifuentes, C. and Sidorov, G., 2026. Knowledge and Context Compression via Question Generation.  [8]     Nonesung, S., Nitarach, N., Jaknamon, T., Taveekitworachai, P. and Pipatanakul, K., 2026. Typhoon OCR: Open Vision-Language Model For Thai Document Extraction. arXiv preprint arXiv:2601.14722.. [9]   Ali, İ., Selim, A. and Skender, F., 2026. Leveraging OCR and AI Tools: A Comparative Guide to Enhancing Data Processing and Decision-Making Efficiency in the Digital Age. Turkish Journal of Engineering, 10(1), pp.129-142.  [10]    Tong, C., Zhang, Q., Li, C., Jiang, L. and Liu, Y., 2026. FaithSCAN: Model-Driven Single-Pass Hallucination Detection for Faithful Visual Question Answering. arXiv preprint arXiv:2601.00269.