Bring new benefits to your business
We developed an initial pilot of DB Lumina. Google Cloud Consultingin the early stages of use case development, I created a simple prototype that only uses embedding without prompts. Although later versions surpassed it, this pilot influenced the subsequent development of DB Lumina’s RAG architecture.
The project then moved through development and application testing environments, into production deployment, and finally went live in September 2024. Currently, DB Lumina is already in the hands of around 5,000 users across Deutsche Bank Research, particularly in sectors such as Investment Banking Origination and Advisory, Fixed Income and Currency. We plan to roll out to more than 10,000 users across corporate banking and other features by the end of the year.
DBLumina is expected to bring significant business benefits to Deutsche Bank, including:
-
Save time: Analysts report significant time savings, saving 30 to 45 minutes preparing earnings note templates and up to two hours creating research reports and roadshow updates.
-
Increased depth of analysis: One analyst increased the analysis of the earnings report by 50%, adding additional sections by region and activity, as well as a summary section on forecast changes. This was accomplished through a summary of earnings announcements and investor notes, followed by analysis with conversational prompts.
-
New analysis opportunities: DB Lumina has created new opportunities for teams to analyze new topics. For example, economic teams in the United States and Europe use DB Lumina to score central bank communications and assess hawkishness and dovishness over time. Another analyst was able to analyze and compare the budget speeches of eight different ministries and aggregate keywords related to capacity constraints and growth orientation to identify shifts in priorities.
-
Improved accuracy: Analysts are also starting to use DB Lumina as part of their editorial process. One supervisory analyst noted that since implementation, there has been a noticeable improvement in editing and grammatical accuracy across notes, especially for non-native English speakers.
Building the future of Gen AI and RAG in finance
We’ve seen the power of RAG transform the way financial institutions handle data. DB Lumina has proven the value of combining search, generative AI, and conversational AI, and this is just the beginning of our journey. We believe the future lies in adopting and refining the “agent” capabilities inherent in architectures. We envision building and coordinating systems in which different components act as agents, all working together to provide intelligent and informed responses to complex financial inquiries.
To support our future vision, we plan to deepen agent specialization within the RAG framework and build agents designed to handle specific types of queries and tasks across compliance, investment strategy, and risk assessment. We would also like to incorporate ReAct (reasoning and acting) paradigm It is integrated into the agent’s decision-making process to enable the agent to not only obtain information but also actively reason, plan actions, and refine searches to provide more accurate and nuanced answers.
Additionally, we will actively explore and implement tools and services available within Vertex AI to further enhance our AI capabilities. This includes exploring other models to accomplish specific tasks and different performance characteristics, optimizing the vector search infrastructure, and leveraging AI pipelines to increase the efficiency and scalability of the overall RAG system.The ultimate goal is to enable DB Lumina to handle increasingly complex and multifaceted queries through improved context understanding, allowing it to accurately interpret context such as previous interactions and underlying financial concepts. This goes beyond answering simple questions and includes providing analysis and recommendations based on the information obtained. To enhance DB Lumina’s ability to provide real-time information and address queries that require up-to-date external data, we plan to integrate grounding response capabilities using Internet-based information.
By focusing on these areas, we aim to transform DB Lumina from a helpful information retrieval tool to a powerful AI agent capable of answering even the most difficult financial inquiries. This creates new opportunities to improve customer service, enhance decision-making, and improve operational efficiency for financial institutions. The future of RAG and gen AI in finance is bright and we are excited to be at the forefront of this innovative technology.
