︎
home     next ︎

molehunt




Molehunt - monitoring skin health using open source AI

Molehunt was created as a demonstration project to showcase the practical applications of Google's open-source MedGemma AI models. The goal was to make a simple and accessible tool, leveraging latest open-source multi-modal AI models, that could be useful for everyday health monitoring.

Key Features:
  • AI-powered mole analysis
  • Interactive 3D body mapping for mole tracking
  • Secure data storage and privacy-first design
  • Professional-grade export functionality for healthcare providers
  • Mobile-optimized responsive design using React and Tailwind CSS  

Tech stack:
  • Frontend: Built with React and TypeScript, styled using Tailwind CSS for a modern, responsive UI.
  • 3D Visualization: Uses React Three Fiber (a React renderer for Three.js) to display and interact with a 3D human model. The model was modified in Blender and exported for web use.
  • Backend: Supabase provides secure authentication, database, and file storage for user moles and logs.
  • AI Analysis: Mole images are analyzed using the medgemma-4b-it model (from Google DeepMind) using custom prompts, deployed on Vertex AI on Google Cloud Platform. The app calls a secure cloud function to run the model and return results.

User flow

Project background:

In July 2025, I saw that Google DeepMind released their new open-source medical model, MedGemma. I wanted to see how it worked, so I decided to build a lil product called Molehunt 🟣.

It's a web app that lets you analyse and track moles on your body using the medgemma-4b-it model. I also had some fun with the UX and built a fully interactive 3D body model (all in React and TypeScript) so you can pinpoint exactly where a mole is and monitor it, with the AI analysing its current state, and any changes over time. I’ve also built a secure sharing + PDF export facility so that you can give these details to healthcare professionals.

Not exactly the sexiest application, but healthcare stuff rarely is. My goal was to build something useful while learning how to take these powerful new models from a research paper to a practical application.

I think it’s interesting that AI could make health tracking more accessible and interactive - mole mapping is not a super accessible thing to do regularly, and tools like this could make it easier for people to stay on top of their health.

Like some of my other recent projects, it was also quite fun to build, and I learnt a lot about 3D web stuff and AI integration using Google’s Vertex AI APIs. For those interested, I put some details about the full tech stack on the ‘About’ page.

Check it out here: https://molehunt.app/