Technical Specification

Platform Architecture & Overview

OculusAI is a personal academic and vision science demonstration project created by Aditya to explore deep transfer learning and interactive optometric screening.

Core Subsystems

Retinal Pathology Engine

MobileNetV2 inverted residual network running 256x256 image tensors to detect Diabetic Retinopathy, Glaucoma, Cataract, and Normal tissue.

  • Transfer-learned feature extractor
  • Circular fundus edge validator
  • Multi-class probability distribution
Ishihara Digit Classifier

Custom 8-layer deep convolutional neural network processing 128x128 plate crops to recognize embedded numerical digits (0-9).

  • 4-block Conv2D + BatchNorm + Dropout
  • 99.52% top-1 validation accuracy
  • Deutan vs Protan likelihood estimation
Vision Deficiency Simulator

Physiological color matrix transformation pipeline calculating real-time spectral absorption variations across all cone classes.

  • Protanopia, Deuteranopia, Tritanopia
  • Cataract optical turbidity filter
  • Dual-view interactive split slider
Digital Acuity Screener

Angular resolution testing utilizing international standard ISO 8596 Tumbling E optotypes with physical display calibration.

  • Standard 5 arcminute resolution
  • Snellen 20/200 down to 20/15
  • Directional response logging

Technology Stack

Interface LayerNext.js 16, React 19, TypeScript, Tailwind CSS v4
Inference BackendPython 3.11, Flask 3.1, Flask-CORS, Pillow
Machine LearningTensorFlow 2.20, Keras 3.x
Document EnginejsPDF, Canvas 2D API
Personal Portfolio Scope

Developed as a personal portfolio project. Models generate experimental probability distributions and vision emulations intended for portfolio presentation and academic study rather than certified medical usage.