Deutsche Bank Research built DB Lumina, an AI-powered research agent that helps analysts automate data analysis and produce research while meeting the strict data privacy requirements of a regulated bank, according to a September 2025 Google Cloud blog post written with Deutsche Bank. Analysts had relied on painstaking manual work, gathering data from financial statements, regulatory filings and industry reports and then synthesising it, which limited the depth and range of their analysis.
DB Lumina has three main features. A chat interface gives analysts access to Google's foundation models, including the multimodal Gemini models, to ask questions, summarise, proofread, translate and draft content, upload and query documents and continue earlier sessions, with guardrails for compliant outputs. Prompt templates provide pre-configured, shareable instructions for tasks such as summarising long documents and extracting key data points, using Gemini's long context window and providing citations. A knowledge feature uses retrieval-augmented generation to ground answers in internal research, external unstructured data such as SEC filings and other repositories, with inline citations, source viewers, controlled access to confidential data and audit logging.
The system runs on Google Kubernetes Engine, with Cloud SQL and the pgvector extension for vectors, Cloud Storage for unstructured data, Dataflow for document ingestion and embedding, Vertex AI with Gemini 2.0 and 2.5, the Discovery Engine API for retrieval and Cloud Natural Language APIs for moderation. Documents are chunked and embedded, and a Python query engine retrieves relevant snippets to give the model context. A custom evaluation framework combines standard metrics with ones such as citation precision and recall and false rejection rates, along with Ragas and human review.
After an early pilot with Google Cloud Consulting, DB Lumina went live in September 2024 and is used by around 5,000 people in Deutsche Bank Research, with plans to reach more than 10,000 users. Analysts reported saving 30 to 45 minutes on earnings note templates and up to two hours on research reports and roadshow updates; one analyst increased the analysis in an earnings report by 50%, and economics teams use it to score central bank communications for hawkishness. "The adoption of the DB Lumina digital assistant by hundreds of research analysts is the culmination of more than 12 months of intense collaboration," said Pam Finelli, Global COO for Investment Research.