Mirae Asset Securities Begins Full-Scale Operation of AWS-Powered 'AI Search for Financial Product Knowledge'
- Input
- 2026-11-20 12:24:47
- Updated
- 2026-11-20 12:24:47

[Financial News] Mirae Asset Securities said on the 20th that it is launching full-scale operations of its “AI Search for Financial Product Knowledge” service, built using generative artificial intelligence (AI) technology from Amazon Web Services (AWS).
The service allows users to search in one place for financial product information, corporate information, disclosures and research materials scattered across multiple systems. When users ask questions in everyday language, AI finds relevant information and provides answers along with sources and as-of dates. The service was designated an innovative financial service in March and deemed suitable in the Financial Security Institute’s generative AI security assessment in July.
Previously, employees had to check multiple systems and documents individually to explain financial products to customers. Mirae Asset Securities expects that being able to ask AI and find the necessary information all at once will improve work efficiency and the accuracy of customer service.
To improve the accuracy and reliability of answers, Mirae Asset Securities designed an “ontology”—a framework for classifying and linking financial product and corporate information—using AWS. It also used “knowledge graph” technology to link about 13,000 documents to product reference data. Multiple AIs were assigned to divide the work of searching for information and drafting answers, and the system was configured to flag information that could not be verified in its holdings as “not held,” rather than generate an answer arbitrarily.
The company also put safeguards in place to ensure security and stability. Amazon Bedrock Guardrails screens for malicious external prompts and harmful content in advance, while Amazon Neptune enables systematic searches of relationships among products, companies and documents. Along with security checks, the company conducted “red teaming” to identify AI vulnerabilities from an attacker’s perspective and established a performance revalidation process to check that answer quality is maintained when models are changed.
The service was built on Amazon Bedrock to support the varied use of “foundation models,” general-purpose AI models pre-trained on vast amounts of data. Frontier models approved for the innovative financial service can be used simultaneously, and improved models can be swapped in immediately after going through the Financial Security Institute’s procedures.
Mirae Asset Securities launched the service for employees last August. By October 20, at least 450 employees had used it to handle more than 2,000 complex queries. Mirae Asset Securities plans to expand the service to customers in the first half of next year.
Mirae Asset Securities also plans to launch an “AI agent for pension operations” next year in collaboration with AWS. The company aims to use AI for pension-related information searches and operational support to improve the accuracy and consistency of customer service and strengthen links with product information.
No Gyeong-hun, head of the financial and public sector at AWS Korea, said, “For generative AI in the financial industry to move beyond experimentation and become a foundation for operations, data structures, security and validation frameworks must be designed together from the outset.” He added, “We implemented an AI service that connects scattered product knowledge and provides sources and as-of dates, after putting it through security validation.”
Ju Se-min, head of the Tech & AI Division at Mirae Asset Securities, said, “This project began with the understanding that data, not models, must be prepared first to put AI to proper use in our work.” Ju added, “Now that we have built a structure in which people and AI can use the product knowledge that had been scattered across systems, it is time to deliver those benefits to our customers as well.”
[email protected] Choi Doo-seon Reporter