During the summer and fall of 2023, I interned as a UX Designer at VoiceXD, where I worked on various features. Among these is the AI-based transcript analysis feature, which was an enhancement to the newly introduced transcript functionality.
Experience strategy and insights, Design, Cross-functional collaboration, UX Writing, Documentation
1 UX Designer, 2 Software Engineers, 1 Product Manager, 1 CUI Design Consultant
8 weeks
We had scoped out certain features as per priority, and it was time to implement other priority features which are crucial for transcript analysis. Using AI for performing the analysis was the top priority. We wanted the users to come to VoiceXD and have a unified experience for building, publishing and analyzing conversational experiences.
Uncovered customer insights with my Manager and defined a priority-based feature stack for VoiceXD. Also handled user queries and suggestions to bring notable updates to the platform.
Translated the business goals and user pain points into effective design ideas, using the existing platform as an anchor. Also designed interactive prototypes based on the user flow for transcript analysis.
Partnered with our Tech Lead to identify constraints and opportunities with OpenAI's APIs to leverage their functionality effectively. Also worked to handle edge cases specific to the APIs responsible for transcript analysis.
Created a reusable component system which can be later integrated into VoiceXD's design system. This made the developer handoff smooth and a new process was established which didn't exist before.
Users need to manually scan the conversation transcripts to extract insights, and scroll all the way up and down to read each message and understand the context of the conversation while being overwhelmed by the information.
Allowing users to use AI to run analysis and refine their Conversational Assistants, from identifying improvement areas in past conversations to gathering actionable insights, all in a collaborative, scalable and efficient manner.
Our backend was designed to consider OpenAI's APIs which had a context limit, and thus if the input exceeds the limit, users wouldn't be able to perform the analysis. Thus, we now have to allow users to select a scope within the conversation to perform the analysis.
We decided to keep the 'Log Analysis' and 'Notes' functionality on the back burner because we didn't receive strong validation regarding their actual usefulness from the users and we were very much focused on the conversation analysis and didn't want to dilute that.
After the user testing, we understood that we had overwhelmed users with multiple features and they wanted a simpler way of performing analysis. Thus, we decided to table the 'Log Analysis' and 'Notes' functionality for later and focus solely on the conversation analysis functionality.
We initially planned to allow users to select a scope withing the conversation to perform the prompt-based analysis, however after user testing, we identified that users might want to see the selected scope at a glance, or quickly reselect the scope. Thus, offering the users with the scope selection functionality in a modal where we planned to offer the run analysis functionality was finalized.
We have now created a space to seamlessly analyze conversation transcripts, without complicating the existing transcripts. Empathy for the win!
VoiceXD could now be used to analyze the conversations between the assistants and people, and gather actionable insights in an efficient manner.
My manager described the design hand-off to developers as "smooth" and "exceeding expectations" considering my role as an intern.
While I harboured an interest in interaction design, and had delved a fair bit into designing for AI, getting my hands dirty raised the bar on my learnings.
Working in sprints in a cross functional team of 4 meant high accountability, speedy decision-making and frequent influx of information to iterate with.
Building feature for users who have limited exposure to CUI design requires a designer to simplify the experience to the last pixel.
How to work with a complex architecture and a non-defined design system to build a solution that is user friendly and effective.
Develop well-defined and informed hypotheses, and then validate them early through in-depth conversations with users and stakeholders.