AI & Machine Learning
DermAI
AI-powered skin disease detection platform
Status
Completed
Industry
Healthcare
Overview
The challenge, in context.
DermAI is an AI-powered skin disease detection platform that helps users and healthcare professionals identify skin conditions directly from medical images. The system performs automated disease classification, generates comprehensive diagnostic reports, and stores a searchable patient history — all through a responsive web interface.
At its core is a pre-trained ResNet-50 backbone fine-tuned through transfer learning, paired with a fast inference pipeline so results come back quickly without sacrificing quality. Every classification is accompanied by a prediction confidence score, letting clinicians judge exactly how much weight to give each result.
The goal was to improve access to preliminary skin analysis while helping medical professionals streamline triage. The result is a system that turns a single photograph into structured, confidence-scored diagnostic information in seconds.
Outcome
DermAI turns a single photograph into structured, confidence-scored diagnostic output in seconds — making preliminary skin analysis accessible to patients and giving clinicians a faster path to triage.
Results & Metrics
Project Scope & Coverage
The numbers and coverage behind the project — measured, reported, and shown here.
Model accuracy during training
Representative convergence of the fine-tuned ResNet-50 across 20 training epochs.
Accuracy stabilises well above the pre-trained baseline.
Dataset split
Standard holdout split used across the curated 262,000+ image dataset.
- Training80%
- Validation10%
- Test10%
What we built
Key Features
AI classification
ResNet-50 powered skin condition classification from uploaded images.
Medical image analysis
Deep-learning analysis tuned for dermatology imagery.
Automated diagnostic reports
Structured, readable reports generated for every analysis.
Patient history management
Past results kept in PostgreSQL for continuity and review.
Confidence scoring
Every prediction shows how much weight it should be given.
Fast inference pipeline
Optimized serving path keeps results quick and responsive.
How it came together
Process
- 01
Discover
Mapped the clinical triage workflow with healthcare stakeholders to define classification scope and reporting needs.
- 02
Data & training
Curated 262,000+ dermatology images across 36 categories and fine-tuned a ResNet-50 backbone via transfer learning.
- 03
Build
Shipped a FastAPI inference service behind a Next.js interface, with PostgreSQL for history and reports.
- 04
Validate
Pressure-tested inference speed, confidence calibration, and report accuracy before handoff.
Stack
Technology stack
Gallery
In context.
A visual look at the project, the industry, and the work behind it.
Have a project like this in mind?