Select Page

Document Classifier and Analysis Tool

Empowering analysts to make better decisions with machine learning and AI

Challenge Statement

Create a custom user interface that enables analysts and data scientists to perform analytics on documents in an air-gapped environment, powered by machine learning and artificial intelligence.

This project was for a federal client. I was the lead designer, working with a team of developers, data scientists, and machine learning experts.

September 2020 – April 2021

IBM Garage, collaboration and outcomes delivered remotely

Sketch, Invision, MURAL, Trello

Our Process

Note: This project was an eight-month, design-intensive effort, but for confidentiality reasons I am unable to share my final wireframes or many details beyond my process.

Design Thinking

We began with a three-day design thinking workshop where my teammates and I led our client through a series of activities to better understand the use case and the challenges our users face. The workshop brought together people from a range of roles to ensure a diversity of ideas and a holistic look at the problem we were trying to solve. My team facilitated several ideation activities and landed on a written MVP statement to guide the delivery phase of our project. 


At IBM Garage, we use a combination of Lean, Agile, and DevOps practices to test our assumptions as we build and deliver value in every iteration. I worked with a team of developers, data scientists, and machine learning experts, along with our client product owner, to discuss user stories and prioritize our backlog every week. 

As the sole designer, I was responsible for the strategy, UX, and visual design of the product. I created user flows, sitemaps, wireframes, and a clickable prototype. I worked closely with the client as a proxy user to pull out user requirements and collaborated with our subject matter experts to design functionality that met those needs. I held frequent design reviews with my developer colleague to ensure my designs were feasible and we collaboratively defined a roadmap for implementing layers of functionality over time. I led a design demo during our weekly playback to present my progress to the client and wider team, taking feedback and iterating on my wireframes as necessary. 

During the early stages of our delivery, I created a task flow to illustrate how our users with different needs would navigate through the tool to accomplish their tasks. The task flow was a great asset to identify areas of complexity and redundancy throughout the project. An early, obscured iteration can be found below.
task flow
When implementing a new feature, I always began by sketching a few possible directions and worked with my wider team to narrow in on a direction. I relied on our researchers and client as proxy users and held design reviews with them to ensure the interface aligned with our users ideal workflow. Once determining a direction for my wireframes, I brought them to a higher fidelity and added them to a clickable prototype for seamless handoff to development. 

wireframes of an graphic interface
Below is an obscured version of one of the core screens of the tool that allows analysts to parse through results with ease. The interface was designed to provide just-enough information at first glance, and has the flexibility of drilling into cards to view more detail.


As our project neared its end, I performed an audit on our tool to make sure our visual and language style remained consistent over our many months of iterations. It was crucial that all parts of this tool felt like a unified, intuitive experience.

I also spent time writing a user guide for the tool. I worked with my team to write the content and created graphics to clearly explain to users how to perform different tasks within the tool. Our finished user guide was a great asset to deliver to our client and was an effective way to showcase all of the functionality within the tool.


For confidentiality reasons, I am unable to discuss the specifics and outcomes of this project beyond saying it was a success and our client was very satisfied. I feel this project was successful in no small part due to the collaboration of design, development, data scientists, subject matter experts, and our client. On a personal level, it was also a great experience in immersing myself in an unfamiliar domain to learn just enough to deliver an intuitive interface that empowers expert users.