Xinyi He

UX Designer
Work  |  Me






GenAI Maintenance Assistant

Conducting two rounds of usability testing before the go-live phase for the MVP GenAI use case: Maintenance Assistant, followed by UX design improvements and a UI redesign to enhance usability.



RoleEmployerTimelineTeam
UX DesignerBMW Group2.5 Months
AI Stakeholders
IT Specialists
Cross-functional UX Partner
Key & End Users







Background



Maintenance technicians had to go through countless documents to find the right information for troubleshooting, making the process slow and tedious.

To simplify this, the Maintenance Assistant was developed as an GenAI chatbot that allowed technicians to ask questions directly and find information within documents more easily. 

The functional prototype was already completed, but before the official launch, usability improvements were needed to ensure the best user experience.




Initial View of the MVP Functional Prototype








Approach



Usability Testing - 1st Round : Functional Prototype

Worked with AI stakeholders and a business IT specialist to test 6 users using the thinking aloud method to identify pain points.

Analysis & Optimization

Used affinity mapping to analyze issues, optimized the chat flow, and redesigned the UI based on the design system for better usability.


Usability Testing - 2nd Round: Redesigned Mockup

Tested the redesigned mockup with 5 users and introduced a cross-functional UX partner to improve broader collaboration and implementation.

Final Outcome

Resolved usability issues within technical constraints, improving clarity and efficiency.




First Round: Functional Prototype, Analyzed and Prioritized Issues
Second Round: Mockup, Deductive Insights for UI & Chat Flow Design





Highlights



A Key Pain Point
Users had to select a database before starting a conversation; otherwise, responses were significantly slow and often inaccurate (AI hallucinations). However, they frequently forgot this step, making the experience frustrating.










Design Process
❌ V1.0

  • Remove the database selection step
  • Provide multiple answer options with relevant documents

--
Result

  • Too many document answers overwhelmed users, causing them to abandon the system early.
  • Context-based default answers were difficult to implement technically.
Design V1.0: Multiple Answer Options After Initial Prompt





Design V2.0: Mandatory Database Selection Before Chat

❌ V2.0

  • Asking users to select the relevant database before each chat

--
Result

  • Relatively complex and tedious




Final Design








  • First-time selection instead of requiring it before every chat, while allowing users to change it at any time later.
  • Default and additional options to balance accuracy and flexibility.
  • Uncertainty option with clear information about potential consequences.
  • Display all database options instead of a dropdown when the selection list is short.


This type of selection challenge is becoming a common design pattern in GenAI products, requiring thoughtful solutions. Beyond databases, it also includes selection for models, information sources, and other key parameters.













Fachportal Redesign
CareCal
Trustworthy GenAI
MusclePal







© 2025 Xinyi He. All Rights Reserved.