01

Qualisense

01

Qualisense

01

Qualisense

QualiSense is an AI-powered quality inspection platform designed to improve defect detection, reduce waste, and enhance manufacturing efficiency. By leveraging adaptive AI and unsupervised learning, it allows manufacturers to train, deploy, and scale automated inspection systems for complex production lines.

Role:

User Research, Interaction, Visual design, Prototyping & Testing

Industry:

AI-powered Quality Control in Industrial Manufacturing

Duration:

2022 – 2024

Overview

At the time, I was working as a Product Designer at TENA Studio, and our task was to design an intuitive, scalable UI that allows factory operators to track defects, analyze inspection results, and refine AI models seamlessly. The challenge was to create a clear and actionable interface that simplifies AI-driven defect detection while integrating into existing factory workflows.

Challenges


One of the biggest challenges was the ambiguity in defect detection. AI models classify defects into multiple categories—scratches, chips, dents, cracks, and other surface imperfections—but these could vary significantly between materials and production lines.

Our goal was to provide clear visualization for defects without overwhelming users. We achieved this by:

  • Categorizing defects with intuitive color coding (e.g., red for critical, yellow for warnings).

  • Displaying confidence levels to help operators validate AI predictions.

  • Ensuring real-time updates so factory workers could act immediately on detected issues.

My Approach

As part of the UX/UI team, I contributed to research, wireframing, prototyping, UI design, and developer handoff. My focus was on creating an intuitive inspection dashboard, streamlining defect classification workflows, and optimizing AI model training interfaces to ensure ease of use for operators.

  • Studied existing quality inspection workflows in manufacturing.

  • Identified user pain points related to defect detection, monitoring, and AI training.

  • Researched how AI-powered automation could improve efficiency while minimizing false positives.

Main Dashboard Layout – Structuring Essential Data


Through research and iterative testing, we determined that operators needed immediate access to key production metrics. The dashboard was structured to provide:

  • A station overview with defect rates, scanned parts, and active cameras.

  • A defect distribution panel, breaking down detected issues by type and frequency.

  • Real-time inspection images, showing current defects for instant validation.

This structure ensured that users could get a high-level summary at a glance while still being able to drill down into detailed inspections when needed.

Factory Hierarchy – Station Overview & Defects Distribution

To create a structured approach, we established a hierarchy model:
➡ Factory → Stations → Cameras



  • Station Overview: Provided a snapshot of scanned parts, defect rates, and active cameras, allowing operators to assess
    system performance quickly.


  • Defect Distribution Panel: Visualized defect frequency and types, making it easier to spot recurring issues across production lines.


This breakdown ensured efficient navigation for operators handling multiple stations across large-scale production environments.

Main Dashboard Layout – Structuring Essential Data


Through research and iterative testing, we determined that operators needed immediate access to key production metrics. The dashboard was structured to provide:

  • A station overview with defect rates, scanned parts, and active cameras.

  • A defect distribution panel, breaking down detected issues by type and frequency.

  • Real-time inspection images, showing current defects for instant validation.

This structure ensured that users could get a high-level summary at a glance while still being able to drill down into detailed inspections when needed.

Factory Hierarchy – Station Overview & Defects Distribution

To create a structured approach, we established a hierarchy model:
➡ Factory → Stations → Cameras



  • Station Overview: Provided a snapshot of scanned parts, defect rates, and active cameras, allowing operators to assess
    system performance quickly.


  • Defect Distribution Panel: Visualized defect frequency and types, making it easier to spot recurring issues across production lines.


This breakdown ensured efficient navigation for operators handling multiple stations across large-scale production environments.

Main Dashboard Layout – Structuring Essential Data


Through research and iterative testing, we determined that operators needed immediate access to key production metrics. The dashboard was structured to provide:

  • A station overview with defect rates, scanned parts, and active cameras.

  • A defect distribution panel, breaking down detected issues by type and frequency.

  • Real-time inspection images, showing current defects for instant validation.

This structure ensured that users could get a high-level summary at a glance while still being able to drill down into detailed inspections when needed.

Factory Hierarchy – Station Overview & Defects Distribution

To create a structured approach, we established a hierarchy model:
➡ Factory → Stations → Cameras



  • Station Overview: Provided a snapshot of scanned parts, defect rates, and active cameras, allowing operators to assess
    system performance quickly.


  • Defect Distribution Panel: Visualized defect frequency and types, making it easier to spot recurring issues across production lines.


This breakdown ensured efficient navigation for operators handling multiple stations across large-scale production environments.

Card Anatomy – Structuring Inspection Results


Each inspection result was presented in a modular card format, ensuring clarity and consistency. Key components included:

  • Defect status (OK, NOK, Error)

  • Inspection timestamp

  • Defect type & severity

  • Highlighted defect region (marked on the scanned image)

  • Camera ID & inspection source

This structured approach allowed for quick decision-making while maintaining detailed traceability of every inspected part.

Card Anatomy – Structuring Inspection Results


Each inspection result was presented in a modular card format, ensuring clarity and consistency. Key components included:

  • Defect status (OK, NOK, Error)

  • Inspection timestamp

  • Defect type & severity

  • Highlighted defect region (marked on the scanned image)

  • Camera ID & inspection source

This structured approach allowed for quick decision-making while maintaining detailed traceability of every inspected part.

Card Anatomy – Structuring Inspection Results


Each inspection result was presented in a modular card format, ensuring clarity and consistency. Key components included:

  • Defect status (OK, NOK, Error)

  • Inspection timestamp

  • Defect type & severity

  • Highlighted defect region (marked on the scanned image)

  • Camera ID & inspection source

This structured approach allowed for quick decision-making while maintaining detailed traceability of every inspected part.

Scalability Challenge – Multi-Camera, Any Aspect Ratio


A major challenge was that QualiSense supports any camera, with varying resolutions and aspect ratios. This required a dynamic grid system that could:

  • Adapt to multiple camera setups per station

  • Ensure uniform defect visualization, regardless of resolution

  • Allow operators to compare images from different angles and depths

We designed a flexible, responsive grid that could dynamically reorganize based on station configurations, ensuring consistent usability across diverse manufacturing setups.


Scalability Challenge – Multi-Camera, Any Aspect Ratio


A major challenge was that QualiSense supports any camera, with varying resolutions and aspect ratios. This required a dynamic grid system that could:

  • Adapt to multiple camera setups per station

  • Ensure uniform defect visualization, regardless of resolution

  • Allow operators to compare images from different angles and depths

We designed a flexible, responsive grid that could dynamically reorganize based on station configurations, ensuring consistent usability across diverse manufacturing setups.


Scalability Challenge – Multi-Camera, Any Aspect Ratio


A major challenge was that QualiSense supports any camera, with varying resolutions and aspect ratios. This required a dynamic grid system that could:

  • Adapt to multiple camera setups per station

  • Ensure uniform defect visualization, regardless of resolution

  • Allow operators to compare images from different angles and depths

We designed a flexible, responsive grid that could dynamically reorganize based on station configurations, ensuring consistent usability across diverse manufacturing setups.


Drill-Down – Single Camera Page

From the station view, operators could select a specific camera to inspect defects in more detail. This page featured:

  • Side-by-side comparison of real-time and historical defect images

  • Zoom and pan capabilities for precise inspection

  • Defect categorization & AI confidence indicators

This level of granularity ensured that operators could validate AI findings and make informed quality control decisions.

Drill-Down – Single Camera Page

From the station view, operators could select a specific camera to inspect defects in more detail. This page featured:

  • Side-by-side comparison of real-time and historical defect images

  • Zoom and pan capabilities for precise inspection

  • Defect categorization & AI confidence indicators

This level of granularity ensured that operators could validate AI findings and make informed quality control decisions.

Drill-Down – Single Camera Page

From the station view, operators could select a specific camera to inspect defects in more detail. This page featured:

  • Side-by-side comparison of real-time and historical defect images

  • Zoom and pan capabilities for precise inspection

  • Defect categorization & AI confidence indicators

This level of granularity ensured that operators could validate AI findings and make informed quality control decisions.

Drill-Down – Archive & Historical Defect Tracking

To improve long-term quality analysis, we developed an archive feature that allowed users to:

  • Search past inspections using filters (date, defect type, camera, station).

  • Compare defect trends over time, helping manufacturers optimize their quality processes.

  • Export reports for compliance and internal assessments.

This feature helped turn defect detection into actionable insights, driving continuous improvements in production efficiency.

Drill-Down – Archive & Historical Defect Tracking

To improve long-term quality analysis, we developed an archive feature that allowed users to:

  • Search past inspections using filters (date, defect type, camera, station).

  • Compare defect trends over time, helping manufacturers optimize their quality processes.

  • Export reports for compliance and internal assessments.

This feature helped turn defect detection into actionable insights, driving continuous improvements in production efficiency.

Drill-Down – Archive & Historical Defect Tracking

To improve long-term quality analysis, we developed an archive feature that allowed users to:

  • Search past inspections using filters (date, defect type, camera, station).

  • Compare defect trends over time, helping manufacturers optimize their quality processes.

  • Export reports for compliance and internal assessments.

This feature helped turn defect detection into actionable insights, driving continuous improvements in production efficiency.

Drill-Down – Model Training Wizard

The AI training process required a step-by-step refinement system that allowed operators to:

  • Mark defect areas manually to enhance AI accuracy.

  • Review and validate AI-detected defects before finalizing training data.

  • Go back and forth between steps, allowing flexibility in refining the AI model.

This iterative training process ensured that AI models improved over time, reducing false positives and enhancing detection precision across various defect types.


Drill-Down – Model Training Wizard

The AI training process required a step-by-step refinement system that allowed operators to:

  • Mark defect areas manually to enhance AI accuracy.

  • Review and validate AI-detected defects before finalizing training data.

  • Go back and forth between steps, allowing flexibility in refining the AI model.

This iterative training process ensured that AI models improved over time, reducing false positives and enhancing detection precision across various defect types.


Drill-Down – Model Training Wizard

The AI training process required a step-by-step refinement system that allowed operators to:

  • Mark defect areas manually to enhance AI accuracy.

  • Review and validate AI-detected defects before finalizing training data.

  • Go back and forth between steps, allowing flexibility in refining the AI model.

This iterative training process ensured that AI models improved over time, reducing false positives and enhancing detection precision across various defect types.


Conclusion

Working on AI-driven defect detection for industrial environments brought several key insights:

  1. Bridging AI Complexity & Factory Usability

    • AI predictions need to be transparent, explainable, and adjustable.

    • Operators must trust the system before relying on automated decision-making.

  2. Creating a Responsive UI in a QT Environment

    • Given the constraints of QT-based development, we had to ensure a lightweight and efficient UI that maintained high performance while handling large datasets.

  3. Scalability & Adaptability Are Essential

    • The system had to support varied camera setups, aspect ratios, and production environments, requiring flexible UI solutions.

This project was an intensive dive into industrial AI applications, ensuring that factory operators could harness AI defect detection without technical barriers—turning AI-driven insights into real-world impact.