SkinDx

Inspiration

People of color are significantly more likely to experience misdiagnosis or delayed diagnosis of skin conditions. One major reason is that many dermatology datasets and training materials are heavily skewed toward lighter skin tones, which causes conditions to appear differently or be overlooked entirely on darker skin.

SkinDx was built to address this gap by leveraging AI to help identify skin conditions more accurately across all skin tones, improving accessibility and equity in early skin health insights.

What it does

SkinDx is an AI-powered skin analysis app that allows users to upload images of their skin and receive real-time feedback on potential conditions. The model analyzes visual features of the image and returns likely skin concerns, helping users better understand what might be happening before seeking professional care.

It is designed to be a supportive tool for awareness and early insight—not a replacement for dermatologists.

How we built it

We built SkinDx as a full-stack web application with an image-upload pipeline connected to an AI vision model trained for dermatological pattern recognition.

We also structured the output layer to return readable insights rather than raw predictions, making results easier for users to understand.

Challenges we ran into

One of the biggest challenges was dealing with bias in training data. Most publicly available dermatology datasets are not balanced across skin tones, which can lead to reduced accuracy for darker skin.

We also struggled with:

• Variability in lighting and image quality from user uploads
  
• Overfitting to certain visual features that don’t generalize well
  
• Balancing accuracy with avoiding overconfident medical claims
  
• Designing outputs that are helpful but not misleading
  

Accomplishments that we're proud of

• Built a working end-to-end AI skin analysis pipeline
  
• Created a system that can process real-world images instead of only curated datasets
  
• Focused the product on health equity and bias in dermatology AI
  
• Successfully turned a complex ML problem into a simple user experience
  
• Delivered fast inference so users get results in seconds
  

What we learned

We learned that building AI for healthcare-adjacent problems is less about raw model accuracy and more about data quality, bias awareness, and responsible communication.

We also learned how important it is to:

• Clearly communicate uncertainty in predictions
  
• Avoid presenting AI outputs as diagnoses
  
• Design for diverse real-world conditions, not just clean datasets
  
• Think critically about fairness in medical AI systems
  

What's next for SkinDx

Next, we want to improve SkinDx in a few key directions:

• Expand and diversify the training dataset across all skin tones
  
• Improve model robustness in low-light and real-world conditions
  
• Add condition tracking over time (progress monitoring)
  
• Introduce dermatologist-reviewed validation for higher trust
  
• Build explainability features so users understand why a prediction was made
  
• Eventually explore clinical partnerships to validate accuracy in real settings
  

The long-term vision is to make SkinDx a trusted early-awareness tool that helps reduce disparities in dermatology access and diagnosis.

Built With

• css
• featherless
• html
• javascript
• python
• vercel