Turning microscopy images into understandable water-quality reports
Turning microscopy images into understandable water-quality reports
AquaIntel is a mobile water-quality monitoring application that combines computer vision and language models to analyze microscopy images and turn the results into understandable reports.
I worked as UI Designer and Front-end Developer, translating a technically complex research system into a mobile interface that makes analysis results easier to understand.
AquaIntel is a mobile water-quality monitoring application that combines computer vision and language models to analyze microscopy images and turn the results into understandable reports.
I worked as UI Designer and Front-end Developer, translating a technically complex research system into a mobile interface that makes analysis results easier to understand.
ROLE
Front-End Developer
UI Designer
STATUS
Completed
2025
TYPE
Academic Capstone
Android
AI
TOOLS
EfficientDet Lite3
SmolLM
SmolLM
ONNX
SmolLM
ONNX
Problem
Problem
Water-quality analysis can produce technical results that aren't immediately easy to interpret.
AquaIntel explores how AI can bridge that gap by turning microscopy analysis into a more understandable reporting experience.
Water-quality analysis can produce technical results that aren't immediately easy to interpret.
AquaIntel explores how AI can bridge that gap by turning microscopy analysis into a more understandable reporting experience.
Solution
Solution
Image → Detection → Assessment → Report
The system identifies microorganisms from microscopy images, processes the results, and generates a natural-language report explaining the findings.
Image → Detection → Assessment → Report
The system identifies microorganisms from microscopy images, processes the results, and generates a natural-language report explaining the findings.
Results
Results
80.7% mAP EfficientDet Lite3's
84.2% recall achieved by the detection model.
90.66% BERTScore semantic similarity score for generated reports
74.7% smaller model
Model quantization reduced the language model's size for mobile deployment.
80.7% mAP EfficientDet Lite3's
84.2% recall achieved by the detection model.
90.66% BERTScore semantic similarity score for generated reports
74.7% smaller model
Model quantization reduced the language model's size for mobile deployment.
FINAL SCREENS
Baby Blue
Baby Blue
#56AEDF
#56AEDF
White
White
#FFFFFF
#FFFFFF
Muted Gray
Muted Gray
#F1F7FA
#F1F7FA
Pitch Black
Pitch Black
#010101
#010101
Pastel Red
Pastel Red
#F6BDC0
#F6BDC0
Mint Green
Mint Green
#C7ECC0
#C7ECC0
Sky Blue
Sky Blue
#B7DDF2
#B7DDF2
Medium Gray
Medium Gray
#767676
#767676
PROCESS
SETTING THE SCENE
The challenge wasn't just detecting microorganisms. It was making the results understandable.
The challenge wasn't just detecting microorganisms. It was making the results understandable.
AquaIntel combines computer vision and a language model to analyze microscopy images and generate water-quality reports.
That meant the interface had to bridge two very different things: technical model outputs and information that a user could actually understand.
AquaIntel combines computer vision and a language model to analyze microscopy images and generate water-quality reports.
That meant the interface had to bridge two very different things: technical model outputs and information that a user could actually understand.
UNDERSTANDING THE SYSTEM
Before designing the interface, I had to understand what the models were actually producing.
Before designing the interface, I had to understand what the models were actually producing.
The application follows a simple pipeline:
The application follows a simple pipeline:
Microscopy image → Detection → Assessment → Report
Microscopy image → Detection → Assessment → Report
Understanding this flow helped determine what information should appear at each stage of the mobile experience.
Understanding this flow helped determine what information should appear at each stage of the mobile experience.
DESIGNING THE EXPERIENCE
Turning technical outputs into a readable mobile experience.
Turning technical outputs into a readable mobile experience.
Instead of presenting users with raw model information, I organized the interface around the questions a user would naturally have:
What did the system detect?
What does it mean?
What should I know about the sample?
This shaped the hierarchy of the results and generated report screens.
Instead of presenting users with raw model information, I organized the interface around the questions a user would naturally have:
What did the system detect?
What does it mean?
What should I know about the sample?
This shaped the hierarchy of the results and generated report screens.
DESIGNING FOR AI-GENERATED INFROMATION
The interface had to explain—not just display.
The interface had to explain—not just display.
The challenge was presenting detection results and generated reports without overwhelming the user with technical information.
I focused on clear hierarchy, readable content, and separating the different stages of the analysis so users could understand the result progressively.
TESTING THE EXPERIENCE
Translating the research system into something people could actually use.
Translating the research system into something people could actually use.
As the UI designer and front-end developer, I worked on turning the research workflow into the Android application.
This meant designing the screens and then implementing the front-end experience alongside the technical system.
As the UI designer and front-end developer, I worked on turning the research workflow into the Android application.
This meant designing the screens and then implementing the front-end experience alongside the technical system.
REFLECTION
Good AI interfaces don't just show the answer—they help users understand it.
Good AI interfaces don't just show the answer—they help users understand it.
AquaIntel taught me that designing around AI is as much about information hierarchy and interpretation as it is about visual design.
The technical system can produce the result, but the interface determines how easily someone can understand what that result actually means.
AquaIntel taught me that designing around AI is as much about information hierarchy and interpretation as it is about visual design.
The technical system can produce the result, but the interface determines how easily someone can understand what that result actually means.