Enhancing Quality,
Reducing Waste & Costs
Enhancing Quality,
Reducing Waste & Costs
Enhancing Quality,
Reducing Waste & Costs
Enhancing Quality,
Reducing Waste & Costs
Up to 99% precision in pharmaceutical visual inspection, lowering costs, reducing waste, improving efficiency & patient safety.
Up to 99% precision in pharmaceutical visual inspection, lowering costs, reducing waste, improving efficiency & patient safety.
Up to 99% precision in pharmaceutical visual inspection, lowering costs, reducing waste, improving efficiency & patient safety.
Up to 99% precision in pharmaceutical visual inspection, lowering costs, reducing waste, improving efficiency & patient safety.
Note: Due to compliance and data protection policies, the designs on my case study have been modified *. To learn more about the product, visit the website.
Note: Due to compliance and data protection policies, the designs on my case study have been modified *.
To learn more about the product, visit the website.
Beyond Labeling: The Bigger Picture
Beyond Labeling: The Bigger Picture
Beyond Labeling: The Bigger Picture
This case study showcases only a small part (Labelling app) of a much larger solution. Achieving high-quality results required collaboration between data scientists, project managers, product owner, engineers, UI/UX designer and domain experts to ensure accuracy at scale. The SaaS solution includes data dashboard, AI model performance analytics, image library, managing application, labelling application, heat maps and much more. Labeling is a crucial step in building reliable training datasets—poor labels lead to ineffective models, limiting inspection performance and increasing costs. To solve this, we designed a tailor-made labeling app that supports labelers in making precise, scalable decisions—ensuring defect-free drug production, AI-driven efficiency, and cost savings.
This case study showcases only a small part (Labelling app) of a much larger solution. Achieving high-quality results required collaboration between data scientists, project managers, product owner, engineers, UI/UX designer and domain experts to ensure accuracy at scale. The SaaS solution includes data dashboard, AI model performance analytics, image library, managing application, labelling application, heat maps and much more. Labeling is a crucial step in building reliable training datasets—poor labels lead to ineffective models, limiting inspection performance and increasing costs. To solve this, we designed a tailor-made labeling app that supports labelers in making precise, scalable decisions—ensuring defect-free drug production, AI-driven efficiency, and cost savings.
Final design*
Final design*
Final design*
Design Principals Applied
Visibility & Feedback
Visibility & Feedback
Visibility & Feedback
Task progress indicator
Clear status indicators for defect detection
Task progress indicator
Clear status indicators for defect detection
Task progress indicator
Clear status indicators for defect detection
Accessibility & Equity
Accessibility & Equity
Accessibility & Equity
Efficiency & Workflow
Efficiency & Workflow
Efficiency & Workflow
Use shortcuts
Predictive inputs to improve speed.
Use shortcuts
Predictive inputs to improve speed.
Use shortcuts
Predictive inputs to improve speed.
*High-Fidelity Prototype – Due to compliance and data protection policies, these visuals and data have been modified for showcasing purposes.
*High-Fidelity Prototype – Due to compliance and data protection policies, these visuals and data have been modified for showcasing purposes.
Project Goal
Project Goal
Project Goal
Designing a labeling system to…
Designing a labeling system to…
Designing a labeling system to…
Accuracy
Accuracy
Accuracy
Enhance quality control by minimizing human error, reducing user fatigue, and improving inspection reliability with AI.
Enhance quality control by minimizing human error, reducing user fatigue, and improving inspection reliability with AI.
Enhance quality control by minimizing human error, reducing user fatigue, and improving inspection reliability with AI.
Safety
Safety
Safety
Enhance defect verification to prevent potential vial content contamination, consequently, minimize patient risk.
Enhance defect verification to prevent potential vial content contamination, consequently, minimize patient risk.
Enhance defect verification to prevent potential vial content contamination, consequently, minimize patient risk.
Simplification
Simplification
Simplification
Reduce the number of images labelers need to analyze by improving AI accuracy and automating repetitive tasks.
Reduce the number of images labelers need to analyze by improving AI accuracy and automating repetitive tasks.
Reduce the number of images labelers need to analyze by improving AI accuracy and automating repetitive tasks.
Waste Reduction
Waste Reduction
Waste Reduction
Improve defect classification through better AI training, minimizing false rejections, reducing waste, and enhancing sustainability
Improve defect classification through better AI training, minimizing false rejections, reducing waste, and enhancing sustainability
Improve defect classification through better AI training, minimizing false rejections, reducing waste, and enhancing sustainability
Research
Research
Research
Problem
Pharmaceutical manufacturers face challenges with visual inspection accuracy
Pharmaceutical manufacturers face challenges with visual inspection accuracy
In the pharmaceutical industry, automated inspection machines with multiple camera stations often misidentify defects, causing two issues:
False Positives: Good products are wrongly ejected, wasting resources.
False Negatives: Defects go undetected, risking safety.
To address this, human inspectors double-check ejected products. However, they often miss micro-defects, making the process unreliable and leading to waste, higher costs, and safety concerns.
In the pharmaceutical industry, automated inspection machines with multiple camera stations often misidentify defects, causing two issues:
False Positives: Good products are wrongly ejected, wasting resources.
False Negatives: Defects go undetected, risking safety.
To address this, human inspectors double-check ejected products. However, they often miss micro-defects, making the process unreliable and leading to waste, higher costs, and safety concerns.
Challenges
Reducing repetitive workload for labelers.
Detecting micro-defects invisible to the human eye.
Improving the precision and consistency of defect labeling to train AI systems.
Reducing repetitive workload for labelers.
Detecting micro-defects invisible to the human eye.
Improving the precision and consistency of defect labeling to train AI systems.
Project overview
Duration: 6 months
Year: 2024
Role: UI/UX
Tools: Figma, Miro, Jira, Confluence
Status: Released
My key responsibilities
Conducting user research
Wireframe and prototyping
A/B and moderated testing
Ensuring compliance and accuracy in the interface.
Team
Stakeholders
Project managers
Product owner
Data scientist
Software engineers
QA | Quality Assurance
User Insights
User Insights
User Insights
To understand user pain points and needs, I conducted an affinity mapping session, grouping insights into four key themes: Human Factors & Cognitive Load, Safety & Compliance, Efficiency, Sustainability and Waste Reduction. Based on these insights, a persona was created to guide design decisions.
To understand user pain points and needs, I conducted an affinity mapping session, grouping insights into four key themes: Human Factors & Cognitive Load, Safety & Compliance, Efficiency, Sustainability and Waste Reduction. Based on these insights, a persona was created to guide design decisions.
Key Insights
01
01
01
Accuracy Challenges
Accuracy Challenges
Accuracy Challenges
Fatigue from high-volume labeling reduces precision, increasing errors and inconsistencies in AI training.
Fatigue from high-volume labeling reduces precision, increasing errors and inconsistencies in AI training.
Fatigue from high-volume labeling reduces precision, increasing errors and inconsistencies in AI training.
02
02
02
Safety & Compliance Risks
Safety & Compliance Risks
Safety & Compliance Risks
Inconsistent defect labeling affects verification, increasing compliance risks and potential safety concerns.
Inconsistent defect labeling affects verification, increasing compliance risks and potential safety concerns.
Inconsistent defect labeling affects verification, increasing compliance risks and potential safety concerns.
03
03
03
Repetitive Workload
Repetitive Workload
Repetitive Workload
Labelers repeatedly classify the same defect types, making the process time-consuming.
Labelers repeatedly classify the same defect types, making the process time-consuming.
Labelers repeatedly classify the same defect types, making the process time-consuming.
04
04
04
Sustainability Concerns
Sustainability Concerns
Sustainability Concerns
High false rejection rates result in unnecessary waste and increased production costs.
High false rejection rates result in unnecessary waste and increased production costs.
High false rejection rates result in unnecessary waste and increased production costs.
Solution definition
Solution definition
Solution definition
Devices
For detailed tasks like defect labeling (zoom, drag, bounding boxes), screen size is critical for usability and accuracy.
For detailed tasks like defect labeling (zoom, drag, bounding boxes), screen size is critical for usability and accuracy.
For detailed tasks like defect labeling (zoom, drag, bounding boxes), screen size is critical for usability and accuracy.
Monitors / Laptops
Monitors / Laptops
Monitors / Laptops
Offer ample space for labeling and multitasking,
with ergonomic setups reducing user strain.
Min. screen size: 13 inches
Offer ample space for labeling and multitasking, with ergonomic setups reducing user strain.
Min. screen size: 13 inches
Offer ample space for labeling and multitasking,
with ergonomic setups reducing user strain.
Min. screen size: 13 inches
Tablet
Tablet
Tablet
Lightweight and portable, ideal for on-the-go work or travel.
Requirement: Stylus/pen for improved accuracy.
Min. screen size: 10.5 inches
Lightweight and portable, ideal for on-the-go work or travel.
Requirement: Stylus/pen for improved accuracy.
Min. screen size: 10.5 inches
Lightweight and portable, ideal for on-the-go work or travel.
Requirement: Stylus/pen for improved accuracy.
Min. screen size: 10.5 inches
Proposed Features
Enhanced Labeling Tools
Enhanced Labeling Tools
Enhanced Labeling Tools
Track progress
Zoom and drag for detailed inspection.
Guideline defects library
Bounding box functionality to highlight defects manually and automatically.
Consistency Checks
Consistency Checks
Consistency Checks
Assigning the same set of images to multiple labels (manager task).
Generating a consistency report with precision scores for each labeler.
Audit trail log
AI-Assisted Labeling
AI-Assisted Labeling
AI-Assisted Labeling
Suggested defect labels with color-coded frames for quick verification.
Editable suggestions to ensure user control.
Design, Iteration & Usability Testing
Design, Iteration & Usability Testing
Design, Iteration & Usability Testing
To assess the prototype's usability and uncover potential pain points, users completed tasks simulating real-world interactions. These tasks targeted key functionalities to ensure the design enhances efficiency, accuracy, and ease of use. This testing phase is crucial in identifying and addressing issues early, preventing costly mistakes in the final design. For feature testing, I prefer using a mid-fidelity prototype over a low-fidelity one, as it provides clearer visual and interactive cues. This helps users better understand their tasks and actions, leading to more accurate and meaningful feedback.
To assess the prototype's usability and uncover potential pain points, users completed tasks simulating real-world interactions. These tasks targeted key functionalities to ensure the design enhances efficiency, accuracy, and ease of use. This testing phase is crucial in identifying and addressing issues early, preventing costly mistakes in the final design. For feature testing, I prefer using a mid-fidelity prototype over a low-fidelity one, as it provides clearer visual and interactive cues. This helps users better understand their tasks and actions, leading to more accurate and meaningful feedback.
To assess the prototype's usability and uncover potential pain points, users completed tasks simulating real-world interactions. These tasks targeted key functionalities to ensure the design enhances efficiency, accuracy, and ease of use. This testing phase is crucial in identifying and addressing issues early, preventing costly mistakes in the final design. For feature testing, I prefer using a mid-fidelity prototype over a low-fidelity one, as it provides clearer visual and interactive cues. This helps users better understand their tasks and actions, leading to more accurate and meaningful feedback.
Test Insights & Design Improvements
Problem
Users struggle to differentiate between defects and selections, leading to confusion and potential errors
Users experienced frustration due to the inability to revise previous labels, increasing the risk of errors.
If an image has a selection but no actual defect, manually marking 'No Defect' each time becomes tedious and inefficient
Added a 'Previous' option, allowing users to revisit and adjust the last three given labels, and a 'Next' option to return to the desired image
Add a "No Defect" button to mark all selections in the image at once, reducing manual effort.
Design solution
Use distinct colors to differentiate defects from selections, ensuring clarity. Make defects stand out over the image by applying a semi-transparent overlay
Problem
Users struggle to differentiate between defects and selections, leading to confusion and potential errors
Users experienced frustration due to the inability to revise previous labels, increasing the risk of errors.
If an image has a selection but no actual defect, manually marking 'No Defect' each time becomes tedious and inefficient
Added a 'Previous' option, allowing users to revisit and adjust the last three given labels, and a 'Next' option to return to the desired image
Add a "No Defect" button to mark all selections in the image at once, reducing manual effort.
Design solution
Use distinct colors to differentiate defects from selections, ensuring clarity. Make defects stand out over the image by applying a semi-transparent overlay
Problem
Users struggle to differentiate between defects and selections, leading to confusion and potential errors
Users experienced frustration due to the inability to revise previous labels, increasing the risk of errors.
If an image has a selection but no actual defect, manually marking 'No Defect' each time becomes tedious and inefficient
Added a 'Previous' option, allowing users to revisit and adjust the last three given labels, and a 'Next' option to return to the desired image
Add a "No Defect" button to mark all selections in the image at once, reducing manual effort.
Design solution
Use distinct colors to differentiate defects from selections, ensuring clarity. Make defects stand out over the image by applying a semi-transparent overlay
Problem
Users struggle to differentiate between defects and selections, leading to confusion and potential errors
Users experienced frustration due to the inability to revise previous labels, increasing the risk of errors.
If an image has a selection but no actual defect, manually marking 'No Defect' each time becomes tedious and inefficient
Added a 'Previous' option, allowing users to revisit and adjust the last three given labels, and a 'Next' option to return to the desired image
Add a "No Defect" button to mark all selections in the image at once, reducing manual effort.
Design solution
Use distinct colors to differentiate defects from selections, ensuring clarity. Make defects stand out over the image by applying a semi-transparent overlay
*Mid-Fidelity Mockup – Due to compliance and data protection policies, these visuals and data have been modified for showcasing purposes.
*Mid-Fidelity Mockup – Due to compliance and data protection policies, these visuals and data have been modified for showcasing purposes.
*Mid-Fidelity Mockup – Due to compliance and data protection policies, these visuals and data have been modified for showcasing purposes.
KPI's
Defect detection accuracy: 99% of defects are now detected by the AI system, compared to manual inspections or legacy machine systems
Task completion time: Reduced by 75%, from 4 minutes to 1 minute per image labeling task after UI an AI improvements (considering the product first version).
User satisfaction: 90% of labelers reported increased satisfaction with the new interface, rating it 4 or higher out of 5.
Lessons Learned
Lessons Learned
Lessons Learned
Key Takeaways
User wellness is the key
User wellness is the key
User wellness is the key
User fatigue directly impacts labeling accuracy and consistency.
User fatigue directly impacts labeling accuracy and consistency.
User fatigue directly impacts labeling accuracy and consistency.
Intuitive Enhancements
Intuitive Enhancements
Intuitive Enhancements
Intuitive design features and AI suggestions can significantly enhance productivity.
Intuitive design features and AI suggestions can significantly enhance productivity.
Intuitive design features and AI suggestions can significantly enhance productivity.
Clear Status Indicators
Clear Status Indicators
Clear Status Indicators
Visual feedback and status indicators are essential for tracking defect detection progress and ensuring user confidence
Visual feedback and status indicators are essential for tracking defect detection progress and ensuring user confidence
Visual feedback and status indicators are essential for tracking defect detection progress and ensuring user confidence
Reflection
This project reinforced the importance of user-centered design in high-stakes industries like pharmaceuticals. By addressing usability challenges such as error correction, visibility, and workflow efficiency, the design ensures a smoother and more accurate defect detection process.
Additionally, incorporating customization options (e.g., light/dark mode, adjustable contrast) highlights the need for accessibility and adaptability in professional tools. The insights gained from usability testing were instrumental in refining the design, proving that small, intuitive enhancements can make a significant impact on user productivity and accuracy.
Moving forward, continuous feedback and iteration will be key to further optimizing the user experience and ensuring that technology supports, rather than hinders, critical decision-making.
This project reinforced the importance of user-centered design in high-stakes industries like pharmaceuticals. By addressing usability challenges such as error correction, visibility, and workflow efficiency, the design ensures a smoother and more accurate defect detection process.
Additionally, incorporating customization options (e.g., light/dark mode, adjustable contrast) highlights the need for accessibility and adaptability in professional tools. The insights gained from usability testing were instrumental in refining the design, proving that small, intuitive enhancements can make a significant impact on user productivity and accuracy.
Moving forward, continuous feedback and iteration will be key to further optimizing the user experience and ensuring that technology supports, rather than hinders, critical decision-making.
This project reinforced the importance of user-centered design in high-stakes industries like pharmaceuticals. By addressing usability challenges such as error correction, visibility, and workflow efficiency, the design ensures a smoother and more accurate defect detection process.
Additionally, incorporating customization options (e.g., light/dark mode, adjustable contrast) highlights the need for accessibility and adaptability in professional tools. The insights gained from usability testing were instrumental in refining the design, proving that small, intuitive enhancements can make a significant impact on user productivity and accuracy.
Moving forward, continuous feedback and iteration will be key to further optimizing the user experience and ensuring that technology supports, rather than hinders, critical decision-making.
Next steps
Leverage these learnings to design scalable, inclusive solutions for other high-compliance industries while continuing to prioritize sustainability and usability.
Leverage these learnings to design scalable, inclusive solutions for other high-compliance industries while continuing to prioritize sustainability and usability.
Leverage these learnings to design scalable, inclusive solutions for other high-compliance industries while continuing to prioritize sustainability and usability.
Portfolio
See other projects
See other projects
Bite - Frozen meal
E-commerce ∙ Case study
KRUU Premium events
E-commerce ∙ Events ∙ UI
AOP The punk rock band
E-commerce ∙ Events ∙ UI
