I led the redesign of an engineer-centered agent training workflow that forced users to complete agent editing before accessing file management.
Instead of following a surface-level fix, I reframed the problem systemically around users' real need to organize data before training. I defined a new UX framework and designed a dedicated file management page within the AI data hub, making local and integrated uploads directly accessible.
This reduced friction, improved workflow efficiency by 123%, and gave users a more independent and intuitive way to manage training data.
Reduced friction in AI training by removing unnecessary decisions and system constraints.
- Removed folder dependency
- Reduced user decisions
- Made training part of the upload process
Flattening a gated workflow into direct access.
The legacy flow required agent configuration before a user could touch files. The redesign exposes file management as a first-class surface inside the data hub, so trainers land where they already want to be.
Multiple cloud services, one account model.
Based on user research, I led the design of a robust multi-account login experience that handles edge cases like token expiration — enabling seamless work/personal account management with a strong focus on the user experience for cross-account file management.
Large file handling, negotiated not assumed.
Previously, large file uploads were not supported, causing failures for files exceeding the token limit. After discussing trade-offs with the engineering team, I introduced minimal UI changes to enable large file uploads and accelerated the development process.
Empowered users with clear interaction.
I led the redesign of the upload status system, closely collaborating with engineers and adapting solutions to cloud provider and API limitations. Now, users receive clear feedback in every upload scenario, reducing confusion and enabling confident next steps.
Native file picker, designed from the docs out.
Read API documentation thoroughly and designed solutions based on technical constraints — led native file picker design and development with consistent behavior across cloud providers and adaptable UX wireframes.
From a learning curve to a landed action.
After redesigning the journey, I introduced intuitive navigation, clear calls to action, and guided upload prompts — eliminating the learning curve and significantly enhancing usability. These improvements drove a 123% increase in feature adoption.
