Oculus3n

Project Summary: Oculus3n

• Problem: Malaysia faces a crisis of preventable blindness due to a lack of specialist access. This gap not only harms patients but also places a heavy administrative burden on our clinicians, who are drowning in referral paperwork and follow-ups instead of focusing on patient care.

• Solution: Oculus3n is an AI platform that unburdens clinicians and empowers community health workers by automating the entire eye screening process.

• Workflow:

1. Register: Instantly scan patient’s MyKad, eliminating duplicate data entry.
2. Capture: Take retinal images with a simple attachment.
3. Analyze: Our AI synthesizes complex data into a clear report in <30 seconds.
4. Act: Immediately identify and refer high-risk patients, automating the referral workflow.

• Impact & Outcome: Drastically increases early detection, reduces healthcare costs, and most importantly, gives clinicians back their most valuable asset: time for patient care.

• Validation: Our core AI model, fine-tuned on a retina focus dataset, achieves >78% accuracy from 52% original MedGemma, giving clinicians high confidence in the automated results.

---

Criteria & Technology

• Alignment with Judging Criteria:
  • Innovation: We decentralize diagnostics, bringing specialist-level screening to the community.
  • User Benefit: Clinicians are freed from administrative tasks and burnout; patients receive faster, sight-saving care.
  • Technical Merit: Our core strength is a custom fine-tuned AI model that reduces cognitive load by providing a clear, actionable diagnosis.
  • Feasibility: The solution leverages affordable, ubiquitous smartphone technology.
  • Deployability: A clear, phased rollout plan ensures sustainable, scalable deployment.
• Technology Stack:
  • AI Model: Google MedGemma, fine-tuned on a private, anonymized kaggle retinal dataset.
  • Capabilities: Screens for Diabetic Retinopathy, Glaucoma, & Cataracts.
  • Cloud & Backend: Microsoft Azure (AI Compute), Supabase (Database).
  • Frontend: Vite + React (Cross-platform Mobile App).

We started with a MedGemma prototype deployed on Azure ML. We are now fine-tuning it with an open-source Kaggle dataset to achieve high-accuracy diabetic retinopathy detection.