Visualizing complex AI data for optimal insights
My Role
As a UX designer working directly with the CEO, CTO, and head of design, I joined the team when the product had an MVP demo and had just finished its first beta testing round. The CEO provided key user feedback, which guided my main tasks of optimizing the MVP demo and enhancing data visualization.
Duration
6 months (2023.9 - 2024.2)
Team
CTO: Talel Lachheb
CEO: Seng Tang
Head of Design: Earl Friedberg
Business Requirement
Optimizing an MLOps platform to lower learning curve and improve accessibility
ML Manager is Daesys's MLOps platform, designed as a B2B AI data monitoring solution. However, the demo has struggled to attract clients and secure partnerships.
Based on client feedback, two primary reasons have emerged as barriers to their willingness to adopt ML Manager:
UX Audit
MVP Demo Covered Main Use Flow, Lacked Accessibility and Learnability
I spent a week conducting a UX audit, summarizing tasks into a brief design handoff for the CEO and Head of Design.
Four key issues were identified that reduced the demo's accessibility and learnability:
Priority Assessment
Four Tasks Prioritization: 2 > 3 > 1 > 4
After meeting with the CEO and Head of Design to discuss the design brief, we aligned our priorities based on severity, ease of implementation, and feedback from beta testing.
How I solve task 2
Designed a non-linear wizard at the beginning
I designed a linear wizard with 5 steps to guide users in creating new tasks, including New APIs, Jobs, and Diagnostics. It is a standard solution that does not require users to learn the cost.
Test & Research
Conversations with target users and dev team
The standard answer doesn't always mean the right one. I did rapid testing with data analysts and got some constructive feedback:
With this insight, I recapped the discussion with the CEO, CTO, and Head of Design to categorize use cases when users create new tasks. I need to learn more about the workflow and necessary information
Revision
Streamlined into 3 steps; Designed progressive disclosure wizard
For creating APIs, Jobs, and Diagnostics, I reorganized the steps based on user actions rather than required information.
For more complex tasks like creating models, datatables, databases, and ML engines, I designed a progressive disclosure wizard—a more optimized solution than a linear wizard to address personalized or specific management needs.
How I solve task 3
Designed an overview dashboard for quick model diagnostics
I extracted model diagnostic information from the raw text format in version 1.0, allowing users to browse key diagnostics through the dashboard quickly. It eliminates the need to click through individual models or rely on email notifications, providing a more efficient way to monitor model performance at a glance.
Quick Fixes
UX Layout Enhancements
To enhance navigation and resolve layout issues, I restructured the 4-level menu into a 3-level navigation, improving the flow of tasks 1 and 4. Additionally, I implemented a slide-in drawer design to address the overlapping area issue, ensuring a more seamless user experience.
3-level Nav Solution :
Slide-in Drawer Solution :
FAQ
More questions about the project that might be helpful
-How do I collaborate with the CEO and CTO as the sole UX designer on the team?
- My collaboration with the CEO and CTO is centered on aligning user experience with both the company’s strategic vision and its technical capabilities.
With the CEO, who usually also takes a PM position in a small team, I aligned that design decisions directly contribute to the business requirements in the weekly meeting and strategically plan the development timeline, ensuring user research and testing were integrated despite limited resources and time constraints.
With the CTO, I work closely to ensure that my designs are scalable and aligned with the technical roadmap. I usually start by collaborating on prioritization, ensuring design and development remain in sync, especially during iterations.
-How do I learn machine learning domain knowledge for the project?
-Two main ways to learn domain knowledge:
Using Knowledge-based AI : I usually use Chatgpt and Perplexity to search for and learn domain knowledge.
Ask the seniors on the team: I usually clean up the questions with the new designs and comment on them in Figma. However, sometimes, for team collaboration, I have to clean them up in a form on Notion/Google Drive.
Discover more
Get in touch
Let's connect on LinkedIn or have coffee chats to explore AI and interactive design together!☕️