Edge Impulse Intern - project dashboard

Project Overview
Edge Impulse is a machine learning platform focuses on B2B service for developers. During my internship there, I redesigned the project dashboard, enhanced the usability by 80%.
Time
10 weeks
Role
Product Design
Research
Team
1 Designer
2 Developers
- Edge Impulse is a machine learning platform
- Users are developers (hardware, software and ML…)
- 65,000+ users
Edge Impulse homepage

About Edge Impulse

We discovered the project dashboard page is not being utilized. The current design lacks a clear overview of project contents, which makes it difficult for users to browse and share projects, ultimately leading to inefficiencies.
    Original version of project dashboard :  complicated and not helpful

    Problem

    Pain points

    Through my heuristic evaluation, I identified 3 key issues that are causing the problem.
    1. Inefficient
    Important information is scattered on multiple tabs and requires users to know what they’re looking for.
    Original user flow to get key info
    2. Indistinguishable
    Most projects look the same from project dashboard, making it hard to distinguish project quality.
    Different projects looks the same
    3. Roadblocking
    Certain project tabs can only be viewed after cloning, create roadblocks for users while browsing.
    Several tabs aren't available
    ‍I looked competitors' dashboard to gain insights, including Github, Hugging Face and Kaggle. My takeaways are:
    ‍
    1. Dashboard provides an overview of a project
    2. Categorize information to help user navigate
    3. Highlight critical details for reference

    Competitive Analysis

    Ideate

    I created 3 versions of the project dashboard, each with a different emphasis and categorization on key info. Through this approach, I aimed to identify the essential information and its hierarchy.
    A. Project overview + data
    Sensor types, project category and output features. Data explorer provides a landscape of the data.
    B. Project overview + impulse
    Target device, sensor type and project category. Impulse section get you quick view of model.
    C. Maximizing information
    Contains most of info within the project. Data section allows you to see train/spilt dataset.
    Later I work closely with my product manager and engineers and iterate from there. We ended up with version C out of 2 reasons:
    • Key information are covered and categorized correctly.
    • Design is flexible and scalable to incorporate different kinds of projects.
    These 2 reasons address the user pain points of inefficient, Indistinguishable. We also address the third pain point by removing the inaccessible tabs.

    Design

    User testing

    I made clickable prototypes and ran the testing with 5 engineers with different focus on machine learning. I facilitate them to use my prototype and learn their thoughts.
    Transcript of UX testing

    General Insights

    Users see the new dashboard helpful and an as huge improvement from the existing design.
    ‍
    “It gives you a high level of flavor of what’s going on. What’s the class of the data, what the question is, sense of where the data is , how well are we doing.”
    “Adding this top level thing is incredibly useful, that’s where you get all of your labels.”
    New project dashboard

    Iteration : Data

    Users wanted to dig in deeper to understand specific data samples.
    ‍
    “ I would have to do a little more digging. What exactly is the data sample size?”
    “Is there a hover? Can I click it?...Super critical when we have visualization is to be able to action in some way.”

    Iteration : Performance

    Users want to be able to compare performance for different devices and optimizations.
    ‍
    ‍“I think it would be helpful for user to see what device is compatible with specific target for this project.”
    “The numbers change depending on the device you choose on the latency calculation.”
    “Definitely want to click it through to see what performance and behavior particularly between quantized and unquantized.”

    Final Results

    80% of users find the new design helpful and a huge improvement. Overall I have..
    ‍- Improve efficiency of project browsing and sharing.
    - Made projects dashboard info more consumable and accessible.

    ‍“There’s good amount of information there. It’s not overwhelming. It’s good balanced.”
    “It’s really useful for me to have the impulse view.It’s really nice to have a highlight of what we’re using.“
    New project dashboard
    New project dashboard + options on performance
    New project dashboard + expanded data explorer

    Takeaway & Reflection

    I'm proud to work with the team at Edge Impulse throughout the summer. It is a complex project and indeed very challenging. Appreciate the team for believing me and supporting me from the very beginning. It’s such a great journey.
    ‍
    Things I learn:
    • When the scope is too broad, make an assumption first
    • Iterate earlier and faster
    • Clickable prototype is really helpful when handing off design to the engineers
    • Try to be as concise as possible when presenting findings
    If I were given more time, I would love to continue to work on dashboard pages to improve the overall design quality includes consistency, discoverability and efficiency for our users!

    Back to home

    © Danny Weng 2022