Waston Discovery Service is a platform which enables developers to apply IBM Watson's various cognitive services to their data, and then their analysts can learn interesting trends and inisghts from it. I've been leading the interaction design work of this project from its initial concept generation to launch. Behind its each release, you would find me in the end-to-end design cycle: researching pain points, brainstorming ideas, storymapping flows, wireframing and prototyping design concepts, usability testing to get feedback, and working with cross-functional teams to get it launched.
I'm going to showcase my work by a few case studies here. Don't hesitate to reach out to learn more! 👻
An important piece of Discovery Service is "building queries" - when users are happy with the ingested and enriched data in our tool, they'd like to explore what they can get from it. Both this concept and the non-SQL query language we use are foreign for the first-time users, so it was a challenge to guide them through this experience. I've spent a good amount of time on improving this experience and testing the changes.
We added a landing page to showcase some interesting results users can get with queries, before getting into the weeds of building them. Starting from each insight card, users can drill down to see deeper insights or modify the query.
We learned that most users prefer to copy and paste a query and modify it, instead of writing a query from scratch. Recognition rather than recall! We cleaned up the copy and order of the query fields, and crafted three help panels to give users a cheat sheet for writing queries.
Fine tuning the interation of the help panels
Iterations of the help panel content
We learned that most of our users started typing natural language questions into the query fields. Well, this was reasonable as we had plain textfields for query inputs. To fix it, we turned them into code boxes to create the affordance of typing code. Better yet, we color coded different components as users and highlighted the correspoinding fields in the result pane, to help them understand the query syntax.
Adding the ability to building queries with a few clicks was a game changer. By providing dropdowns and populating values, we lowered the barrier for creating queries to a new extent. Usability tests suggested that this feature was especially helpful for complex query function like Aggregation. Users can also see a live update of the query language as they make selections, to further help them understand and learn the query language.
Another challenge we had when designing Watson Discovery Service was: the data users uploaded has converted into a structured format, but there wasn't an easy way for them to know what exactly the format is. We had iterations of ideas for solving this, including some fancy visualizations, but eventually this "reviewing a list of data fields and nested fields" concept tested well. it comes with two views:
The collection view lets users see and explore all the top level and nested fields across all documents in their collection. They can toggle through some example values of a field to get an idea of what the field is about. We also provide in-context sample queries for each field, to enable users to make their first query using this field.
In usability testing of the collection view, one consistent feedback we got was users wanted to drill down to a specific document and check its structure, to help them understand what's different now comparing to the original document. Therefore, we designed this document view that allows users to explore all the enrichments applied to each of their documents in a single view.
Another project I worked on was Watson Knowledge Studio. Knowledge Studio provides a comprehensive tool set for human/machine annotation works. It does all the behind-the-scenes work to make the Watson magic happen. I paired with a user researcher to ideate, design and create this product from scratch - it was a very challenging and rewarding journey. It gave me the chance to experience a whole UX design cycle from user research all the way to high-fidelity design. Its technical complexity and contraints around diffrent user roles made the design process extra fun.
We did a two-week on-site contextual inquery to learn about and synthesized four user roles that are involved in the ground truth generating process. Below is one example of the four personas we created.
In IBM design thinking, "Hill" is a high level Epic with a practical user-centric goal. Due to the complexity of this new project, we needed to decompose them into a set of scenarios to help us, engineers and other stakeholders understand them. Therefore, inspired by Karen Holtzblatt's contextual design model, I created two visualizations called "Day in the Hill" for each Hill, and shared among engineers, PMs and our target internal users. It turned out to be very helpful for making a complex workflow simpler.
Hill 1 foused on illustrating how a Subject Matter Expert and a training enginner can work together to create a POC annotator within one week.
"Day in the Hill" - Hill 1 visualization
Hill 2 focused on illustrating the iterative process of human annotators annotating documents and reaching consensus.
"Day in the Hill" - Hill 2 visualization
One challenge was designing UI for the multiple roles using the same tool. With their different needs in mind, I designed a stremlined and flexible workflow for the Process managers, and a very focused view for human annotators who just need to work on annotating documents.
Landing page for Process manager
A calendar view of the scheduled manual annotation tasks, for Process manager
A process manager can check the status of each set of the assigned tasks.
Human annotator's view for annotating documents
As the only designer on the team by then, I also worked on the visual design of this project. I took the chance to define the style guide for future designers to use.