Data Labeling Platform
Our data labeling tool MVP for Microsoft helps users quickly tag and categorize images, text, and other data so it can be used effectively in machine learning models.
Data labeling involves marking or annotating data with relevant information—like identifying objects in an image or key phrases in text—to train AI systems. Our tool speeds up this process with customizable templates, reducing manual effort and improving efficiency, especially for large-scale projects. It simplifies workflows and helps machine learning models get accurate, organized data faster.
SUMMARY
Project Overview
We designed and launched the MVP for the platform, delivering six key features: three primary features (image, text, and search labeling) and three secondary features (user, content, and training management). A centralized dashboard was developed to streamline user and task workflows, enhancing the overall user experience.
We built a comprehensive design system from scratch, defining all essential UI components such as buttons, form fields, sidebars, and error states to ensure consistency and scalability across the product.
Collaborating closely with front-end engineers, we conducted design QA to ensure pixel to pixel implementation.
The MVP was engineered to scale, supporting 10,000+ users while maintaining high performance and usability.
Project Timeline
UNDERSTAND
How does data labeling work and what are the current applications?
USER INTERVIEWS
What do users struggle most with in the current process?
To gather insights, we conducted in-depth interviews with doctors trained to label scan results and screened AI products to explore how data labeling is used in the medical field—one of our key end users for our product.
We visited Dr. Pei and Dr. Huang from the Thoracic Surgery Department of Beijing Haidian Hospital to interview them based on their experience. Through this process, we have a clearer understanding of the significance of making such products. Also, we were able to bring some valuable insights to the table during the meetings.
Dr. Pei has about 10000 minutes of experience in manually labeling CT scans of lung cancer. Given that our AI labeling software is designed to automate and assist in data labeling, Dr. Pei’s explains his understanding of the pain points, challenges, and necessary precision in medical labeling.
Read the full interview here.
USER PERSONA
Who’s will be using our product?
Once moving to the interface and styling, we developed a user persona—derived from our affinity maps—to better understand key usage scenarios and the core needs of our users. This persona helped ensure that the UI team (Yixing, Ivy, and Zhe Li) aligned on a shared understanding of our target user.
We took a close look at how task creators (judges) and task executors (labelers) interact with the system, identifying key pain points and opportunities to improve task assignment and streamline the overall labeling process.
COMPETITIVE ANALYSIS
Who are the key players in the field, and what solutions and features do they currently offer?
Our team started the project by looking at key players in the data labeling space to help shape our product. By comparing features, design, and ux, we’re focusing on areas where we can stand out and improve what’s already out there.
Through our research, we found that Labelbox has a feature set that’s pretty aligned with what we’re aiming for. Conversations with our client and stakeholders also highlighted them as a major competitor. Based on their product, we identified ways to simplify the user experience for new users while keeping the advanced functionalities we need, like real-time label previews and support for different formats like images, text, and search labels.
AFFINITY MAPPING WORKSHOP
Prioritizing Features for MVP
After gathering insights from competitive analysis, user research, and stakeholder meetings, we organized an evaluative workshop with PMs, designers, and clients. The goal was to assess and prioritize features while discussing their technical feasibility to move forward in the design process.
We created an affinity map to define the features for the MVP, identifying three main features and three ancillary features.
Information Architecture
Based on the affinity mapping workshop, we created the information architecture map below, outlining the defined features and their supporting elements.
IDEATE
User Flow
Based on feature list defined above and competitive analysis done, we worked with the product managers to define the tasks flows below for content management, training management, and image labelling flow.
Click on the images to view in detail.
IDEATE
Design System
Since we were designing and building the MVP for this project, there was no existing design system in place. The client preferred not to use Microsoft’s design system, so we developed three proposals and ultimately chose the one outlined below. Since we haven't established the product's branding yet, and the client might select a different primary color in the future, opting for a single color is a more cautious and flexible approach.
IDEATE
How did we visualize the experience?
After defining the features, we began visualizing the user experience for the labeling template creation process and dashboard layouts. We reviewed Microsoft’s internal product, Skillrise, along with competitor solutions, and iterated based on both their designs and our client’s needs to enhance usability and streamline workflows.
DESIGN
How did we design the system to be intuitive?
Below shows the annotated UI for the label template creation process.
While we recognize that judges are already familiar with creating templates, our goal was to provide a modular and intuitive interface that further simplifies their workflow. We have annotated key features developed based on both user and client needs. One notable example is the customizable drag-and-drop template creation tool, which allows users to easily tailor their search templates to specific requirements.
Search Labeling
Text Labeling
Image Labeling
DESIGN VALIDATION
How did we validate and track design changes for accuracy during development?
Given the tight development timeline, the front-end engineers began development as we were still working on the high-fidelity prototype. This allowed us, as designers, to conduct timely walkthroughs and identify areas for improvement. To streamline the review process and ensure clear communication, we set up a shared Excel sheet between the design and front-end teams to track page modifications. This enabled us to provide timely feedback and synchronize updates after each revision.
Final Product Demonstration
NEXT STEPS
Implementing AI Assisted Labelling for 2.0
In our next steps, we will implement AI-assisted labeling utilizing large language models to optimize the labeling process for images, text, and search data.
Additionally, we are developing advanced features, including image labeling tools such as segmentation masks, as well as support for video labeling. These updates for Beta will provide users with more robust capabilities for data annotation and analysis.
1. Design is never ending.
At iSoftstone, we constantly revisited and iterated on the design because we knew that during affinity mapping, we had only prioritized certain features for the MVP, but we needed more to stand out in the market. The way I built it in a scalable manner allowed the team to pivot and adapt for the long term.
Although we didn’t get the chance to conduct testing for the MVP, we’re now prioritizing user and usability testing for the Beta version. This feedback is crucial for refining the product, addressing pain points, and improving the overall user experience. We're also integrating LLMs to enhance data labeling and adding new features like video labeling. We've drafted usability testing questions to gather insights on both the technical functionality and how well the platform supports labeling tasks.
2. Design is about collaboration.
A designer’s role goes beyond just creating visually appealing designs; it's about collaborating to ensure those designs are practical and can be successfully implemented. At iSoftstone, with the support of my team and managers, I was able to tackle the challenge of enhancing AI data labeling workflows in just 4 months. hey are an incredibly talented group, and working alongside them was a great experience. I'm grateful for everything I learned, and their collaboration was key to achieving this success in such a short time.
3. Design is about how things work.
Design is not just about how things look; it is also about how they work. As a designer, it is my responsibility to ensure that everything I do at this company prioritizes creating designs that can be built, will be built, and align with the customer's tone.
4. Turning design into full business products.
Although I’ve designed B2B products before, working on the AI data labeling project gave me a clearer perspective on how design truly supports a product team. It deepened my understanding of the challenges we’re solving and showed me the importance of collaborating with different stakeholders to tackle those issues. It’s been rewarding to see how design evolves into fully developed business solutions and how it helps simplify complex processes.