Loading...
Perplexity BirdSQL
The Re-Designing of
What is Perplexity BirdSQL?
Perplexity BirdSQL is an AI powered search engine specifically focusing on the database of Twitter. It can translate natural human language input into SQL, a program language which is widely used by professional data scientists, and then the SQL would single out the data of users' interest.
Our Design Challenge
Our challenge is to tailor this product to a specific user group who may not be familiar with the website but could benefit from it. Seeing the potential of how the search engine can be used by data scientists -- either beginner or masters -- to accurately collect information about particular topics, we decided to figure out how BirdSQL can be more useful for this group of users.
Chapter 1: pain points
User interview & observation
After interviewing 5 people who all possessed certain level of data science knowledge and haven't heard of BirdSQL before, we identified some key findings:
- users don't have a clear idea of what the page is for: what's the dataset?
- users have trouble creating an original search question because they don't know what kind of questions are allowed to ask -- they usually started exploring the page by clicking some prompt questions and then customize the prompts
- users with intermediate level of data science knowledge may not understand SQL language
- users with advanced level of data science knowledge could understand SQL language, but the code section is useless because they can't interact with the code
- users couldn't export the search result in a form that can later be easily analyzed using other expert data science tools
User personas
After organizing the problems users encounter we created three personas to remind us that we should emphasize with our intended user group as we were starting to think about the solutions.
Problem Statement
People with a certain level of data science knowledge who are not familiar with Perplexity BirdSQL need to have more flexibility customizing search results in order to collect data of their research interests about certain cultural trends on Twitter more effectively.
Chapter 2: the solutions
Now it's the ideation part! We came up with a couple of new features surrounding our problem statement in our low-fi prototype and we designed two versions of each user flow interface. The features can be categorized into 3 aspects: the main page section, the coding section, and the visualization section.
1. The main page
- moving the introduction of BirdSQL to a location that is more obvious than the original website
- classify the prompt search question so that users can have a better sense what kind of questions they can ask
- added gear icon in each prompt allowing users to directly customize the prompt and then start searching from there
2. The coding section
- allow users to directly make changes on certain part of the code and run the customized code
- allow users who are not familiar with SQL code to also engage with the code section by either providing them other program languages or providing brief explanation about the SQL keywords if they hover certain terms they don't understand
Version A
Version B
3. The visual section
- allow users to view other forms of visualization of the result data
- if the result is loading for a very long time allow user to terminate the process (so that they can start a new search right away)
- allow users to export the data so that they can use it later for further analysis using other data science tools
- allow users to add a new search page on the same page so that they can compare two data at the same time
Chapter 3: user testing
Key findings:
"But why do I need that when I already can open another tab on my browser to search another question?"
- surprisingly, users didn't find the add-search-page function very beneficial because if they really want to compare multiple data results they can use their browser to open a new window for a new search
- users didn't find the add-search-page function very beneficial because if they really want to compare multiple data results they can use their browser to open a new window for a new search
- Users prefer a simple interface layout without too many add-on features
- users think having access to alternative visualization forms can be helpful but they are all confused by the term "alt form" at first
Chapter 4: High-fi showcase
Taking users feedback seriously, we deleted some features while modified others. Overall speaking bellow are the demonstrations of all the revised new features, which still are divided into 3 categories:
1. The main page
classification of search prompts
customization of search prompts
2. The coding section
alternative program language + customization of codes
3. The visual section
stop running when loading time is too long
get back to previous search using history function
alternative data visualization + data export
Chapter 5: User testing 2.0 & revision
Although we were more confident about our design once we made our high-fi prototype, we still wanted to further improve on our new features as well as the layout of our new features. Consequently, we picked three interfaces which we deemed as the interfaces needed adjustments the most to conduct another round of user interviews. Afterward we made several key revisions based on the feedback we got. Below are the before-after stories of these 3 UIs.
Code Section Layout 2.0
Since a lot of users expressed the desire to have the code section ready at hands all the time as they scroll through the main result section, we make the code part sticky on the top of the screen as soon as the visual field no longer covers the code section.
Search History 2.0
Since most users disliked the bordered form of the history section, we replaced it with simplistic horizontal lines and we also changed the color of the text into a lighter one so that it won’t compete with the vibrant blueness of the title “History”. Besides, we added details of search history in terms of the specific changes users make for the code, enabling users to get back to a particular version of the question they are interested in.
Data visualization 2.0
Given that all users did not like the horizontal scrolling style of switching to other data representation forms, we made some additional space for statically displaying this kind of information so that users can see what other options are available right away.
Key lesson:
What looks like a great re-design idea from designers' perspective is not necessarily aligned with what users truly value!