Project Overview

Project Summary:

Our service is designed to prevent the increasing and serious problem of fraud in Korea's predominant housing form, the Jeonse lease contract. Unlike other existing services, we have created a service where AI assesses the risk of Jeonse fraud with simple document reviews. Additionally, we have built a chatbot service capable of accurate Q&A by applying RAG technology to LLM, a generative AI. RAG technology was implemented using LangChain and Vector DB, to reduce LLM's hallucination issues.

Identifying the Challenge

The Social Problem:

Jeonse fraud in South Korea, which exceeded 4 trillion won in damage in 2023 and even resulted in suicides, is a grave social issue. This problem leads to financial losses and contributes to distrust in the real estate market. Additionally, it undermines housing stability and causes community unrest. Addressing these issues is crucial for fostering fair and stable living environments, thereby enhancing the stability and credibility of local communities.

Innovation and Uniqueness

Why Our Project Stands Out:

Unlike existing services that solely offer document review feature through real estate agents, resulting in long waiting times and high costs, this service provide real-time document review and Q&A features using AI in addition. Additionally, we review all documents necessary for preventing Jeonse fraud. Lastly, we included an online community feature to allow users to continue engaging with the service even after utilizing the Jeonse fraud detection feature.

Insights and Development

Learning Journey:

Our team faced the challenge of dealing with the latest technologies, LLM and RAG. We engaged with relevant research papers to accumulate theoretical knowledge. The unexpected discovery we made was recognizing the benefits of studying technologies through academic papers. We realized that learning through research papers is an effective method to conduct a deep analysis of the topics.

Development Process:

We developed an algorithm detecting the risk of Jeonse fraud and a Q&A system using LLM. We integrated RAG into LLM to enhance accuracy and minimize hallucination issues. We integrated these technologies into a website and implemented online community features. We ensured clear role allocation among team members to maximize their expertise and regularly shared updates to align project direction.