# Customized FinGPT Search Agents Using Foundation Models

**Authors:** Felix Tian, Ajay Byadgi, Daniel Kim, Daochen Zha, Matt White, Kairong Xiao, Xiao-Yang Liu Yanglet  
**arXiv:** [2410.15284](https://arxiv.org/abs/2410.15284)  
**Date:** 2024-10-20  
**Summary page:** https://www.summarizepaper.com/en/paper/2410.15284v1/

## Key points

- Authors present a paper titled "Customized FinGPT Search Agents Using Foundation Models" at the 5th ACM International Conference on AI in Finance in 2024.
- The paper addresses limitations of existing large language models (LLMs) like BloombergGPT and FinGPT in catering to specific user needs within the financial domain.
- FinGPT Search Agents are customized for two distinct user groups: individuals and institutions to overcome shortcomings of current models.
- For individual users, Retrieval-Augmented Generation (RAG) techniques are used to integrate local documents and user-specified data sources into the search process, enhancing relevance and accuracy of results tailored to individual preferences.
- For institutional users, dynamic vector databases and fine-tuning models on proprietary data are utilized to meet specialized needs of organizations operating in finance.
- Development considers challenges such as data privacy concerns, time-sensitive nature of financial information dissemination, and swift response times.
- Rigorous experimentation shows that these tailored agents outperform existing models across metrics like accuracy, relevance, and response time.
- The research contributes significantly to enhancing the utility of language models in analyzing financial information by offering tailored solutions for diverse user requirements within both individual and institutional contexts.

## Summary

Authors Felix Tian, Ajay Byadgi, Daniel Kim, Daochen Zha, Matt White, Kairong Xiao, and Xiao-Yang Liu Yanglet present a paper titled "Customized FinGPT Search Agents Using Foundation Models" at the 5th ACM International Conference on AI in Finance in 2024. The paper addresses the limitations of existing large language models (LLMs) like BloombergGPT and FinGPT in catering to specific user needs within the financial domain. To bridge this gap, the authors introduce FinGPT Search Agents customized for two distinct user groups: individuals and institutions. The  are designed to overcome the shortcomings of current , such as BloombergGPT and FinGPT, when it comes to meeting the unique requirements of users in the . This is achieved through a tailored approach that caters to both individual and institutional users. For individual users, the authors utilize Retrieval-Augmented Generation (RAG) techniques to seamlessly integrate local documents and user-specified data sources into the search process. This personalized approach enhances the relevance and accuracy of search results tailored to individual preferences and requirements. On the other hand, for institutional users, dynamic vector databases are utilized alongside fine-tuning models on proprietary data to ensure that the search agents meet the specialized needs of organizations operating in the . This ensures that these agents are highly effective in providing timely access to relevant financial information. The development of these customized FinGPT Search Agents takes into consideration key challenges such as data privacy concerns, time-sensitive nature of financial information dissemination, and swift response times. Through rigorous experimentation and evaluation, it is demonstrated that these tailored agents outperform existing models across metrics like accuracy, relevance, and response time. As a result, they are deemed highly practical for real-world applications within financial institutions and among individual users seeking efficient access to pertinent financial data. In conclusion, this research contributes significantly to enhancing the utility of language models in analyzing financial information by offering tailored solutions that cater specifically to diverse user requirements within both individual and institutional contexts. The refined approach presented in this paper showcases promising advancements in leveraging  for developing highly effective search agents in finance.