Franziska Lorenz

Average reading time: 6 minutes (1009 words)

The team

I worked along the project with a freelance UX designer and many frontend-, backend- and platform developers. It was supported by the research department, quality assurance department and a number of stakeholders.

My tasks

My main tasks were the following:

interview stakeholders and accommodate their wishes
analyze and understand the current tool
identify and restructure important parts
establish workflow for users without experience
create detail concepts
enhance the style guide
support the implementation

Starting the project

The starting point was the internally used platform CWB 1.0, consisting of a purely technical user interface.

User Interface of the internally used tool Cognitive Workbench

One of the most important moments of this project was the first stakeholder kick-off and interviews. We made sure that all existing stakeholders were involved in the process and agreed on the priorities of the project.

Pain Points

Workshop, individual interviews and discussions with stakeholders revealed some pain points:

required types input types are not clear
no apparent workflow
user is limited to one data source combined with one kind of analysis
naming of the elements is too difficult
design was not optimized for the user experience (instead optimized for power users)
no view to provide an overview of the analysis
not possible to run several analyzes in parallel
UI doesn't show which components need configuration
no user and rights management

Key things to redesign

clear presentation of the workflow to getsuccessful results
combining sources and analysis methods should be variable to try different things
allow the user to revoke and revise decisions at any time
configuration workflow and data exploration should be separated features
both features should be quickly accessible
share results with other people
everyone should get an own user account with dedicated rights

Research

To gain inspiration, we concentrated considering systems with focus on configurability of a data flow and presentation of many complex contents (i.e. Apple Quartz Composer, Prototyping tools). To make decisions about our strategy we als looked at competing systems like IBM Watson. We observed that visual programming languages are often used as solution for configuration interfaces. They provide a good overview of the overall structure and at the same time can be changed quickly and easily. In addition, they work at an extremely high level of abstraction.

Competing systems and other inspiration sources

Analysis

As a starting point for the analysis, we had the system explained to us by the QA staff, collected all the components and linked them. This analysis should help to understand all components when it comes to re-sorting and expansion.

Breaking down and defining the currently available content

Definition of content and networking components

The analysis confirmed our initial idea to subdivide the content into configuration and exploration.
Configuration
Allows the import of data from external systems or other users. The data can be linked to various NLP algorithms, combined and represented in a component named “Thinker”.
Some of standard algorithms are trained to recognise e.g. people, organisations or the relation between them.
Dataset
By linking the thinkers to the imported data, the algorithms can access and analyze the data.
To provide maximum configurability, multiple data sources and Thinkers can be created and linked. Results to be shown are determined by the inputs/thinkers connected to the Dataset.
Exploration
An exploration mode can be used to query the Dataset to result in different representations. A possible form of representation is the display of an imported document with the extracted entities that the algorithms have found in the text.

Content splitting into Configuration and Exploration - both parts have access to the Dataset

Configuration mode: Consisting of data input and Thinkers using NLP algorithms to extract information

Visualization: Each Dataset can be displayed in a wide set of output methods and compared with each other

Exploration Mode: In addition to comparing different Datasets, it is also possible to explore the data for different representations

Sharing & Export: Results can be shared with other users

Wireframing

The components developed in the analysis phase were integrated into the new structure. The resulting information architecture served as basis for additional features like creating analyses in parallel.

Information architecture: Compilation of all contents and linking of logical sequences

For the configuration of an analysis (in the following: scenario) we have decided for a kind of visual programming language as a suitable paradigm.

Brainstorm and visualize initial ideas on how the contents could be linked as simply as possible

To get feedback stakeholders and users, we used short explanatory videos to visualize the general concept. Using an Invision prototype, they were able to test the system for a few days and evaluate the idea.

Different ideas for maximum freedom in the configuration interface

Concept

The processing of the feedback was already done in high fidelity concepts. At the same time, the developers began restructuring the backend. Implementation and conception were continued in parallel.

Hi-fidelity screendesign delivered to the implementation team

To extend basic concept and turn the modular system into a product, we created detailed concepts for a dashboard, a navigation flow and configuration interfaces. At the same time, we developed a simple style guide that contains color and typography definitions as well as individual components in order to obtain a uniform design and quickly put together the visual design.

Problems

There were a variety of problems during the project: the first problem was the very limited time and little programming experience of such complex user interfaces.
We had no user base or actually defined target group, which made the whole research part close to impossible. There were no experienced designers on board and the design team was too small for such a task in time.
In addition, other small problems have emerged, like the chosen framework for the connection lines, which was not flexible enough and difficult to adapt.

My ideal process

The first problem was the large scale of the project combined with limited time and little experience developing highly complex user interfaces. There were no experienced designers on board and the design team was too small for the task at hand.
We had no user base or defined target group, which made the whole research very difficult.
We also faced a number of smaller problems during the implementation e.g. the very rigid framework for the connection lines causing lots of effort to be adapted..

Side projects

share concept between users
documentation of the system for delivery to the customery
integration of a help center
creation of legal pages as additional HTML pages
revision of almost all components and development of a glossary

Portfolio

About

Say hi

Construction kit for AI applications

The project