DenizBank and Intertech wanted to move away from a manual, workstation-based approach to AI and ML model development. Previously, each model required its own complex setup, with separate environment variables, database connections, and code stored locally on individual workbenches. This made it hard to standardize, monitor, or reuse work, and slowed down time-to-market.

To address this, they adopted Red Hat OpenShift AI on top of their existing Red Hat OpenShift foundation. The goal was to create a comprehensive, standardized, self-service model development environment that:

• Automates data science pipelines
• Provides consistent standards for naming, workbench creation, and resource usage
• Scales model serving more easily
• Improves operational efficiency and cost control

With OpenShift AI, more than 120+ data scientists across risk, marketing, and customer relations now work in a shared, standardized platform. They can:

• Spin up tailored workbenches using pre-built or custom images (for example, GPU-enabled Python images)
• Use integrated tools like Jupyter Notebooks, TensorFlow, and PyTorch
• Rely on GitOps practices so environments are defined as code and can be destroyed and rebuilt quickly

Intertech, supported by Red Hat Consulting, designed an architecture with 15 OpenShift AI clusters on bare metal, on-premise. This setup gives data scientists a self-service, standards-based environment while IT maintains governance and alignment with DevOps and GitOps best practices.

By adopting Red Hat OpenShift AI, DenizBank and Intertech have been able to reimagine how AI models are built, validated, and deployed, with several measurable benefits:

• Greater autonomy for data scientists: More than 120 data scientists from different lines of business now use a self-service platform to create their own model development environments. They no longer depend on manual environment setup by IT for each new model.
• Consistent standards: Intertech defined clear standards for naming conventions, workbench creation, compute and GPU requirements, and pipelines. This reduces variability between projects and improves visibility into each model’s full environment.
• Faster time-to-market: Intertech expects the time to provision a new model environment to drop from about 1 week to roughly 10 minutes, thanks to automation and self-service deployment pipelines.
• More robust and secure models: Code is now stored in a central repository, enabling code reviews and better validation. Guardrails ensure data scientists follow standards and regulations, even as they work more independently.
• Improved operational efficiency: OpenShift AI integrates model serving, data science pipelines, and notebook environments, which simplifies the end-to-end AI/ML workflow and reduces manual effort.

Leaders at DenizBank and Intertech highlight that OpenShift AI provides a streamlined, AI-focused environment that supports both innovation and governance. It helps them build more reliable models for use cases like credit risk prediction and fraud detection, while keeping development cycles shorter and more predictable.

DenizBank and Intertech use Red Hat OpenShift AI to make GPU and compute resources more efficient and predictable for AI workloads.

Key elements of their approach include:

• GPU-aware base images: Data scientists can choose pre-built cluster images, including GPU-enabled Python images, so they get the right libraries and hardware acceleration out of the box.
• GPU slicing and scaling: OpenShift AI integrates with NVIDIA dashboards to monitor GPU usage and automatically adjust the size of GPU slices allocated to each model as needed.
• NVIDIA Multi-Instance GPU (MIG) technology: Through the NVIDIA GPU Operator, GPUs are partitioned into multiple isolated instances. Each instance has its own compute, memory, and bandwidth, allowing multiple pods across different projects to share GPU resources with:
• Guaranteed quality of service
• Fault isolation between workloads
• Higher utilization without extra hardware: This approach maximizes resource utilization and lets more AI workloads run simultaneously, without immediately investing in additional GPU hardware.

Beyond GPUs, Intertech is also focusing on improving overall compute resource utilization (CPUs, GPUs, and memory) and further enhancing model inferencing performance on OpenShift AI. The combination of GitOps, standardized templates, and GPU slicing helps them run AI workloads more efficiently while maintaining flexibility for different teams and projects.