Authors:
Liu Yuqiao, S. Rubin Bose, J. Angelin Jeba, Cao Yu, Kobayashi Shota, Takahashi Daiki

Addresses:
Department of Information and Communication Engineering, The University of Tokyo, Bunkyo, Tokyo, Japan. School of Computer Science and Engineering, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India. Department of Electronics and Communication Engineering, S.A. Engineering College, Chennai, Tamil Nadu, India. Department of Computing and Information Systems, Singapore Management University, 81 Victoria Street, Singapore. Department of Information Science, Osaka University, Suita, Osaka, Japan. Department of Engineering, University of Tokyo, Bunkyo, Tokyo, Japan.

Abstract:

Information overload plagues financial analysts today. Equity researchers tracking even a handful of businesses see thousands of news stories, earnings releases, analyst comments, and macroeconomic updates each trading day. Hand-sifting through this amount for meaningful insights is time-consuming and increasingly impossible. For fast-paced research, current AI technologies lack real-time data or require substantial human programming. This study introduces a Retrieval-Augmented Generation (RAG) pipeline that lets equity analysts have natural-language discussions with a controlled stream of daily financial news about individual businesses. Researchers import daily articles from Reuters, Bloomberg, and Yahoo Finance; chunk and embed them using the all-MiniLM-L6-v2 sentence transformer; store them in an FAISS vector index; and query them with GPT-4. A purpose-built prompt-engineering method keeps the model in the recovered context, reducing hallucinations. Our evaluation covers six months of financial news (January to June 2024) for Tesla (TSLA) and Apple (AAPL), totalling 12,000 articles. Our hybrid RAG system achieves precision of 0.89 and NDCG of 0.91, significantly higher than BM25 baselines. Experts gave our method a 4.71/5.0 fidelity grade for generation quality. The suggested pipeline shows that RAG-based systems can be reliable, practical research assistants when carefully tuned for finance. A domain-specific RAG methodology, a curated financial QA evaluation benchmark, and analyst-facing application prompt engineering insights are presented in this study.

Keywords: Retrieval-Augmented Generation; Financial NLP; Equity Research; Conversational AI; Vector Embeddings; Financial News Analysis; Large Language Models.

Received on: 29/11/2025, Revised on: 12/01/2026, Accepted on: 03/03/2026, Published on: 16/10/2026

DOI: 10.69888/FTSFDS.2026.000801

FMDB Transactions on Sustainable Finance and Data Science, 2026 Vol. 1 No. 4, Pages: 213-222

• Views : 53
• Downloads : 7

Download PDF