DEVLOG #2
Productive day — got the core ML pipeline working and built out the web interface:
-
Trained the detection model — fine-tuned EfficientNet-B0 on 80k+ retinal OCT images. Hit ~95% validation accuracy across 4 disease classes (CNV, DME, DRUSEN, NORMAL). Training took ~25 min on Colab’s GPU. Model weights saved and ready for Pi deployment.
-
Built prototype web interface — simple site for device usage and real-time results display. Users can see detection confidence scores + disease info. Planning to add more educational content about each condition. You can try it here: veya-web-zeta.vercel.app
Next: quantize the model down to ~5MB for faster inference on Pi, then test the full pipeline (capture → predict → display) end-to-end.
Comments 3
bro i aspire to have your intelligence
hahahaa thank you so much broski, hearing it from the goat itself ))
super cool!!
Sign in to join the conversation.