Kore Geosystems uses AI to help geologists work faster. The scanning, uploading, cropping and preparation of rock core tray images are part of their Ingest system (image management).
The product team wanted to convert their Windows app to the web so people could access it from anywhere and work together more easily. They also needed ways for users to upload images from different sources like DSLR cameras and other systems. I was tasked with streamlining the rock scanning and analysis workflow to improve geologist user experience.
When I first came into the team, I had to learn how core (rocks and minerals) analysis affects geologists' work within the mining field. Our team included several subject matter experts who regularly shared insights into their complex work.
While our AI accelerates analysis (Logging), the image scanning and management process introduces new challenges for geologists who previously never dealt with digitizing rock images. My role was to ensure this workflow ran seamlessly with AI.
Receiving customer feedback
Even though our in-house geologists are not direct users, they frequently received customer input about issues with our legacy Windows application, which they then communicated to me.
Interface & Navigation Issues
Users find the interface unintuitive, requiring explanation from geologists and support staff
The overall contrast is weak, making links and buttons obscure
Some fields do not have enough context or information
Wayfinding, navigation and visual context of the boxes (core trays) are lacking
Operators need to keep switching screens when reviewing scanned images
"I'm completely lost with these settings - it's my first time scanning a core tray."
Workflow & Process Problems
Operators often enter wrong box numbers or swap box orders while scanning
Mistakes aren't noticed until images are already cropped and organized (days/weeks later)
Operators feel they need to take on more responsibilities, slowing productivity
Tedious organization and search for specific images
Many editing features are only available through backend support (requiring calls/emails)
"No. When you have hundreds of images, scrolling through them to find a particular image is time-consuming"
"How am I supposed to make a new box when I don't even know what number to give it?"
"Can't I just scan this thing and let the system handle the box numbering?"
Technical Limitations
Images take time to load after switching pages
Not enough variation of image quality/renders for different needs (reviewing, zooming, comparing)
Operators can't figure out proper cropping or templating of boxes
The templating has strict rules users may not know beforehand
User Anxiety & Expectations
Users are afraid to export images for logging because actions can't be undone
Users expected AI assistance during the process
Desire to import images from other software or archives
Some parameters, such as padding or box stack direction are not relevant for geologists.
“I don’t understand what these fields mean.”
"Let me know if there are any mistakes or if things are overlapping."
Brainstorming ideas with the team
Since our original application contained no AI assistance for image cropping, I facilitated a series of brainstorming meetings to establish our strategy. Throughout these discussions, I emphasized customer pain points and feedback.
User workflow
When we implemented AI to support image cropping and management functions, our team needed to reevaluate our entire technical architecture. Although complete automation was the target, data resource challenges required us to deploy the solution in phases.
Image & data structure
Since we were getting images from more places than our own scanners, I discussed with the team about all the different image types and compositions we might see. I wanted to cover all the edge cases so our support team would know what to expect.
Sketching
After getting all the must-have details from the devs and SMEs, I created a table view that lets users see the big picture - showing each hole (project) and how it connects to other parts of the app.
Old state
Our initial workflow was too rigid, leading to high dropout rates among new customers. They struggled to troubleshoot errors in the product, forcing us to spend resources helping them.
Current state
We've enhanced our workflow to fix troubleshooting issues and improve cross-platform accessibility. While AI automation saves significant time for operators, lingering legacy requirements contribute to product bloat.
Future state
Our goal is to lift data restrictions on images, so users can focus on quality. Enhance AI cropping capabilities and create a comprehensive image hub where users can perform all actions without navigating excessive features.
Usability testing conflicting design ideas
While striving to provide customers with the best user experience possible, we simultaneously needed to respect timeline constraints and deliver the product according to schedule. Our testing of various design options led to a solution that effectively balanced usability considerations with resource efficiency. The engineering team and I chose MUI framework to streamline design and development.
We wanted to give customers a great experience, but we also had to stick to our timeline and deliver on schedule. After testing different design options, we found a solution that balanced usability with efficiency.
Even with our framework, design changes were needed for better UX. The left design (original framework) puts the "new image" button above parameters, while the right places it at the bottom for smoother transitions.
I wanted to see which table information was most helpful for users. Even with images in the grid, participants still completed tasks successfully, though about 20% slower than without images. It's worth considering that including these images could significantly impact page loading performance.
Ingest cam
Our scanner app has been renamed Ingest Cam and now supports DSLR cameras for greater accessibility. The UI has been streamlined to focus on two key tasks for camera operators: capturing quality images and viewing all photos in a single hub.
Improved image capture
Previously, customers struggled with poor visual context, low contrast, and excessive data entry for core tray photos. The new design makes it easier to take photos one after another while reducing data requirements.
New image management hub
Our old cluttered page is now a customizable hub where operators can fix mistakes easily. These improvements make troubleshooting more intuitive, reducing support team workload.
Ingest image management (web)
The Ingest Web management hub now features AI-powered image cropping, helping users prioritize image quality and sequencing.
Mirrored image management hub
The image hub is also available on the web platform with more functionality that supports imports from third-party applications and local storage.
AI-cropped image view
Users can also review or edit AI-cropped images and override any errors.
Updated image cropping tool
The new cropping feature offers greater flexibility by enabling users to adjust multiple x and y coordinates for complex core trays.
Importing images from other sources
Despite the new import feature improving accessibility to our product, it still requires numerous data fields inherited from our legacy architecture.
In November 2023, Ingest Cam and an improved Ingest image management (web) were launched during a mining conference, boosting customer satisfaction and generating more detailed feedback from our users.
Support Ticket Reduction
Reduced related monthly support ticket volume from around 50 to 20 tickets
Feature Adoption Usage
Over 80% of projects successfully used our AI crop tool within 3 months of release.
Project Image Uploads
Increased image handling 10x (tens to hundreds of GB/project) in a year
Future and thoughts
Operators hit a learning curve when switching to this product since regular core logging doesn't usually utilize image systems. Looking ahead, we need to improve three main areas: AI tools, cropping options, and ways to edit image info. This would make everything run smoother and speed up core logging. However, these improvements require significant investment in scalability, data structure modifications and AI development, which means solutions to current user needs may not be immediate.
Working in a startup, I came to appreciate product design more deeply during this project, as I realized that better user experience often requires greater investment of time and resources to bring to market. Finding the right balance involved multiple discussions with the team.
