Back to WorkTianyu Wu

2024

Raymics On-Prem & Federated Learning

Extending the SaaS design system to hospital deployment and multi-institution collaboration

A 0–1 B2B platform for department directors and IT — multi-role permissions on-premises, plus federated learning across institutions without moving raw data.

Role
UX Designer
Team
Product · research, workflow, demo, and design-system reuse
Timeline
2024
Scope
· Smooth interaction and design concepts supported million-level hospital contracts · Batch operations improved efficiency and reduced errors · Reused SaaS components to shorten development
Raymics Fusion on-premises and federated learning platform hero

Overview

Hospitals needed local deployment for data control, and multi-institution research needed federated learning so models could train without sharing raw patient data. Building on Raymics Cloud, I designed the on-prem research platform and federated collaboration flows as one enterprise story — not two disconnected products.

Complex stakeholder system

Directors create cooperation projects and open permissions; doctors participate and work with public or shared data; information departments monitor authority changes; Fusion nodes accept desensitized uploads from each institution and return a shared model. The UX had to make roles, permissions, and data paths legible.

Complex stakeholder system across on-prem, Fusion nodes, and permissions

IA & workflows

Information architecture covered data upload and checks, research topics and tasks, AI model training, public-data permissions, and setup — with a process spine from upload through training. The on-prem front page kept Cloud interaction patterns while adding denser data visualization for hospital operators.

Information architecture wireframes across five main navigation processes
On-premises front page with stats, usage flow, and quick entries
Role permissions page for apply, audit, and access management
AI model training UI from prep through aggregate results

Design system reuse

Card sizes, navigation, and status colors stayed aligned with Cloud so users could adapt quickly. Differentiation came through specialized data cards, federated institution pickers, and batch operations for doctors running multi-source training.

Design system reuse and consistency across SaaS surfaces

On-prem & federated results

Locally deployed products supported million-level hospital contracts. Batch processing improved operation efficiency and reduced errors. Reusing Cloud components shortened development.

Feedback and key results for on-prem and federated products

Supporting product lines

Beyond on-prem and federated learning, the suite also needed coherent desktop tools, community, and website surfaces — so researchers could move between products without relearning the system.

Supporting product lines hero for desktop tools, community, and website
Medical data processing tool overview: position, cores, and references
Build frames for navigation hierarchy and scalable dashboard layout
Landing page design comparing card versus table display
Prototype collage of dashboard, forms, and modules
Main process workflow from import through ROI, desensitize, and upload

Medical community

We designed a 0–1 medical community with a clear information architecture, consistent grids and navigation, and social interactions that connected back to the product suite.

Medical community 0–1 information architecture
Community design detail with 3-column grid and consistent top nav
Community hover states, social actions, and cross-product switching
Editing, post, and save-draft process for community content
AI tool-assisted design workflow with Cursor, v0, and Claude
Launched community pages on desktop and mobile

Official website

The official site IA and launched pages — home, products, pricing, and mega-menu — presented the enterprise suite as one coherent brand story.

Official website information architecture
Official website launched pages: home, products, pricing, and mega-menu