Project & Award: Developed during the 1st Baekgyeong Hackathon (Digital Smart Busan Academy), where our team (‘뉴스 요약해조’) received the Excellence Award for the service CATCH! NEWSFIN.
Figure 1. Framework of Chatbot with RAG
“The No. 1 problem in today’s generation and economy is the lack of financial literacy.” – Alan Greenspan
In highly volatile markets, accurate information processing is critical. This project tackled the widespread issue of low financial literacy and information asymmetry by engineering CATCH! NEWSFIN, an automated, LLM-driven financial news curation service.
📌 The Problem: Financial Illiteracy and Info Overload
Every day, the market is flooded with complex financial news. Retail investors and the general public often lack the financial literacy (average score of 66.5 in Korea) to filter through the noise, summarize core impacts, and extract actionable insights. This cognitive overload leads to a significant disadvantage in asset management and economic decision-making.
⚙️ The Method: Local LLM Hosting & RAG Pipeline
To empower individual decision-making, we designed an intelligent, multi-faceted news processing architecture utilizing Retrieval-Augmented Generation (RAG):
- Automated Scraping & Vectorization: Built a pipeline to scrape daily articles from the Maeil Business Newspaper. Using the OpenAI
text-embedding-3-largemodel, we converted these articles into vector representations and stored them in a Vector DB. - Local Hosting & Model Validation: Implemented LangChain and LangServe to construct the RAG system. For the core generative AI, we deployed the
yanolja/EEVE-Korean-10.8B-v1.0model hosted locally via LM Studio. We validated its superior inference and scenario-generation capabilities by benchmarking it against thellama-3-Korean-Bllossom-8Bmodel. - Core Services: The architecture powers three main features:
- Daily News PICK: 1-line summaries of the previous day’s most popular news.
- Financial Perspective Interpretation: AI-driven analysis of how specific news impacts other sectors like real estate and stock markets.
- Interactive AI Chatbot: A LangServe-based playground where users can actively ask questions about selected articles.
🚀 The Impact: Enhancing Cognitive Decision-Making
This project successfully transformed overwhelming, unstructured textual data into structured, easily digestible intelligence. By anchoring the generative AI to actual news data via RAG and utilizing a highly optimized local LLM, CATCH! NEWSFIN bridges the information gap, reduces cognitive load, and serves as a powerful tool to elevate the financial literacy of retail investors.