Background and requirements

Stakeholders wanted an MVP product of a large language model/AI assistant to be utilized across the Sharp network by its employees. They needed this quickly due to the privacy and PHI concerns involved specific to the healthcare industry. This began as a developer-led project but early on in the process, I was pulled into the process along with a product manager to help optimize the user flow and patient experience.

• General AI assistant available for people to use, with some reduced capabilities (e.g. no image creation or coding assistance)
• Receive feedback from users
• Branding

• Comprehensive privacy policy and terms of use
• Meeting transcription summarization tool (scope added part-way through)

Meeting with stakeholders

We were met with many challenges in this area as there were already ideas and expectations from stakeholders given I was pulled into the project after kick-off. One particular area the required a lot of work on our part to convince stakeholders was regarding the display of some key navigation items they wanted to highlight on the landing page.

Stakeholders had communicated desire for items like privacy policies and FAQs be highlighted in the final product as shown in the blue ovals on the right. From a UX perspective, this was visually distracting and detracted from the main action intended for this page, which is to use the AI assistant.

Competitor research

I researched different existing AI engines and interfaces out there and it was very clear these interfaces showed streamlined designs where the focus on the screen was the question and answer components. This helped us make a case to bring to stakeholders.

Advocating for the design

• We presented a clean, streamlined design where navigation items were folded into the menu
• Navigation items were still discoverable and in an intuitive location
• Stakeholders were eventually agreeable to the design solution we offered

Scope was expanded

Over halfway through this project, stakeholders wanted to add a new feature to our tool - a teams transcription summarization tool. This posed several new challenges for a team.
‍
New challenges:

• Our legal team was very wary of both teams recordings and transcriptions so there was a lot of back and forth for our terms of use
• Unexpected design and development work
• Another couple of sprints had to be added to accommodate for this new feature and we had to advocate for our launch date to be pushed back

‍A few of the many added design requirements:

• Queue management: ensure that the queue for uploaded files is efficiently managed, and that each file’s processing status is accurately tracked.
• Status transition and error states: Clear transition from one state to another (e.g. uploading to summarizing) as well as states indicating any user or system error.
• Notification system: the system should notify the corresponding user-facing components via a reliable messaging system or service hooks.

Design iterations

Iteration #1

I explored a modal layout for this tool as it seemed like an easy way for users to switch between the chat feature and the summarization feature. However, I moved in a new direction for several reasons. 1) Users using the summarization feature may not necessarily also be wanting to switch tasks between the chat. 2) There could potentially be a lot of interaction with this tool, which may not be appropriate for a modal.

Second iteration

The next iteration moved the summarization tool to its own page. I initially split the page into a top section (uploading) and a bottom section (summarizing and downloading completed tasks). I also thought it would be a nice experience to be given a time estimate of the summary time. Ultimately, it still felt like there was too many moving pieces happening at once. Also, a time estimate ended up being out of scope for the team at this time.

Final iteration

In the final iteration, all transitioning and active tasks in one space (upload, waiting in queue, and summarizing) occur at the top of the page and then a second section below if for everything that was completed and ready for download. This made it more clearcut for the user what is currently happening and where to download finished documents. This design accounted for all the requirements and it was easy and intuitive to use.

SharpAI