ML Model Management Portal

Interviews were conducted with stakeholders across multiple roles:
• Project Managers (Hydrography)
• Geophysicists (Seismic, RAMMS)
• Innovation Team Leads
• Solution Owners (Offshore Wind)

Key Insights from Interviews:
• “We often don’t know why the system fails or which data causes errors.”
• “Model retraining requests go through multiple technical teams — it’s slow.”
• “We need visibility on who changed what, when, and why.”
• “Sometimes retraining locally gives better results than relying on global models.”

These insights informed the creation of features like automated validation testing, data traceability, and multi-level model management (local, generalized, and global).

I used an Affinity Diagram to organize insights from user interviews and observations into six key themes — Data Upload & Preparation, Model Training & Testing, Model Management, Integration & Collaboration, Error Management & Iteration, and End User Operations. These clusters highlighted user challenges across the ML workflow and guided the creation of personas and core product features.

Based on insights from interviews and shadowing sessions, I developed personas representing key user roles — such as ML Engineers, Project Managers, and Geophysicists. These personas captured their goals, frustrations, and workflows, helping maintain a clear user-centered focus throughout the design process. They also guided feature prioritization and ensured that each design decision aligned with real user needs.

I developed an Information Architecture (IA) in collaboration with the Product Owner to define the structure and hierarchy of the ML Center platform. The IA organized complex workflows — including data management, model training, and validation — into a clear and intuitive navigation system. This blueprint ensured logical task grouping, minimized cognitive load, and supported seamless transitions across the platform’s modules.

The ML team lead initially developed a preliminary wireframe before the discovery phase, which served as a starting point for the project. I refined this concept into low-fidelity wireframes, incorporating insights from user research and the IA to improve workflow clarity, feature placement, and task efficiency. I also drew inspiration from the original wireframe while iterating with stakeholders and Subject-Matter Experts, validating core interactions and ensuring the design met real user needs before progressing to mid-fidelity prototypes.

Building on the low-fidelity wireframes, I developed mid-fidelity wireframes to add more detail and structure to the user interface while maintaining focus on usability and workflow. These wireframes incorporated visual hierarchy, interactions, and data presentation elements, reflecting insights from user research and feedback from subject-matter experts. I also drew inspiration from the ML team lead’s initial wireframe, refining layouts and navigation to better support model management, validation, and collaboration tasks. This stage allowed for further iteration and alignment before moving into high-fidelity designs.