Deutsche Bank, like many global banks, remains vigilant about data security and compliance. Although they have a close partnership with GCP as part of their cloud-first strategy, much of their sensitive data still resides on-premises.
This has led to a hybrid cloud model, balancing innovation in AI and Machine Learning with strict confidentiality requirements. The challenge lies in deploying cutting-edge solutions without compromising security standards. As a result, teams must juggle regulatory obligations, complex tech stacks, and a constant push for modernization.
Hybrid AI Platform
The AI Hybrid Platform was created to keep pace with the bank’s cloud-first strategy, while addressing the sensitive nature of on-prem data.
By blending OpenShift 4 on-premises with Google Cloud services, it reconciled the need for innovation with strict compliance requirements. Terraform-driven infrastructure-as-code and automated CI/CD pipelines were implemented to maintain strong security controls, essential for handling sensitive data.
Thousands of users now have a streamlined way to engage with AI models and share insights across the organization.
I led and developed the end-to-end development of this platform, focusing on seamless integration between on-prem systems and GCP.
I collaborated with analytics teams, soon-to-be AI teams, as well as security specialists and leadership to ensure each layer of the design met the bank’s robust standards. From designing the Terraform scripts to setting up CI/CD workflows, developing the Helm Charts, I orchestrated a solution that is both scalable and easily maintainable. I also implemented best practices that make the platform portable, ready for adaptation to other banking divisions in the future.
Keywords:
Kubernetes, MLOps, Cloud Computing, Openshift4, GCP, Terraform, Helm, Docker, IT Security, Data Security
Deutsche Bank’s AuditGPT project aims to streamline the work of auditors, who navigate a maze of regulations and constantly evolving rules. By leveraging a Retrieval-Augmented Generation (RAG) pipeline, it brings together internal guidelines, past audits, and other compliance data to answer questions in real time.
The system uses open-source frameworks like Ollama for model serving and OpenWebUI for a custom user interface, enabling auditors to securely interact with generative AI features.
Models from Hugging Face were integrated to ensure flexibility and performance, all while hosting the solution primarily on-premises. This approach respects the bank’s stringent data confidentiality requirements and mitigates concerns over third-party cloud usage.
I played a pivotal role in designing the pipeline, setting up the RAG workflows, and curating the knowledge base.
I also configured Ollama to effectively load and run the open-source models, while integrating OpenWebUI to provide auditors with a user-friendly interface.
My efforts ensured that all regulatory guidelines and past audit data were seamlessly accessible through an AI-driven assistant. By meticulously addressing security measures and compliance needs, I helped deliver a robust, future-proof solution for auditing challenges.
Keywords:
GenAI, Kubernetes, RAG, Ollama, Open WebUI, Data Security, Models, On-Prem
PDS – Payment Data Service is a large-scale initiative at Deutsche Bank aimed at consolidating payments data across multiple regions and legal entities into a single, unified format.
Historically, each branch operated its own system, creating a fragmented data landscape that hampered analysis. By migrating this consolidated data to Google Cloud Platform (GCP), PDS seeks to deliver both security and flexibility.
The goal is to enable the Analytics Team to build cutting-edge models and solutions within a secure analytical workspace that taps into the bank’s vast payments data.
This environment also supports collaborative efforts and makes it easier to share insights and results. One of the initial proofs of concept involved detecting potentially fraudulent transactions and identifying payments that circumvent existing regulations.
I was instrumental in extending the existing PDS architecture, ensuring that the data flows and security protocols met the highest standards. Working closely with the Analytics Team, I designed an environment that allowed them to explore and build AI solutions without compromising sensitive data.
I also established workflows for PoC testing, enabling analysts to quickly iterate on potential fraud-detection models. By defining clear governance policies and implementing secure access controls, I helped guarantee that compliance remained intact at every stage
Keywords:
Cloud Computing, Big Data, GCP, Kubernetes, Jupyter Hub, BigQuery, Data Governance
