Diabetes is a global challenge, and Diabetic Retinopathy remains a leading cause of preventable blindness. For many, the first symptom is permanent vision loss. SightX was built with a personal mission: to bridge the gap between advanced medical AI and the patients who need it most, honoring a family journey with diabetes.
By combining a ResNet-50 V2 backbone with Bayesian decision theory and a "No-Line" clinical UI, SightX provides a robust, safe, and beautiful screening experience.
SightX is built as a highly-decoupled microservices stack, ensuring scalability and clinical reliability.
📖 Read the Detailed System Design & Architecture Document ➔
(Includes details on our RHEL institutional deployment model, security flow, and data lifecycles).
| Component | Responsibility | Tech Stack | Documentation |
|---|---|---|---|
| Frontend | Clinical Interface & User Flow | React, MUI, Vite | View README |
| Backend | API Orchestration & Gateway | Node.js, Express, Multer | View README |
| Inference Engine | ResNet AI & Clinical Post-processing | PyTorch, FastAPI, NumPy | View README |
- 108-Iteration TTA Ensemble: Robustness against camera artifacts using Test-Time Augmentation.
- Bayesian Prior Correction: Adjusts for training-set bias (EyePACS) to reflect real-world clinical prevalence.
- Risk-Minimized Decisions: Uses an asymmetric cost matrix to prioritize patient safety over raw accuracy.
- The "No-Line" Rule: A design philosophy utilizing tonal layering and depth instead of harsh borders.
- Clinical Aesthetics: Glassmorphism and high-performance animations tailored for medical environments.
Before launching the SightX stack, ensure you have the following installed:
- Docker & Docker Compose: (Required for Orchestration)
- Install Docker Desktop (includes Compose).
- Supabase Account: (Required for Persistence)
- Local Dev Tools (Optional - only if running services individually):
- Node.js 18+: For Frontend and Backend.
- Python 3.10+: For the Inference Engine.
The entire SightX stack is containerized for professional deployment.
- Environment Config: Ensure your
.envcontains the required Supabase and AI engine tokens. - Launch Stack:
docker-compose up --build
- Internal Access:
- Frontend:
http://localhost:80 - Backend API:
http://localhost:5001 - Inference Engine:
http://localhost:8000
- Frontend:
SightX uses Supabase (Postgres + Auth + Row Level Security) to manage practitioner accounts, clinical roles, and diagnostic records.
The complete Supabase setup guide — including project creation, schema SQL, RLS policies, superuser bootstrapping, clinician registration, and troubleshooting — lives in the Backend README:
Quick overview of what's covered:
| Step | Description |
|---|---|
| 1–3 | Create Supabase project, grab API keys, configure .env |
| 4 | Disable email confirmation for local dev |
| 5–6 | Run schema SQL (tables, trigger) and RLS policies |
| 7 | Create the first superuser via the Supabase dashboard |
| 8 | Register clinicians through the app's admin panel |
| 9 | Log in as a clinician and access the dashboard |
Tip
If you're setting up SightX for the first time, start with the Backend README — it has the full step-by-step walkthrough.
SightX is optimized for high-resolution retinal imaging. To ensure diagnostic accuracy, images must be captured using professional Digital Fundus Cameras:
- Field of View (FOV): Minimum 45° horizontal (non-mydriatic preferred).
- Resolution: Minimum 30 pixels per degree (ppd).
- Standards: Images should ideally be DICOM-compliant with unique patient identifiers at the point of capture.
- Environment: Controlled lighting to minimize artifacts and lens flare.
SightX is designed for institutional deployment and adheres to strict clinical governance:
- Environment: Must operate exclusively within Hospital/Clinical Environments under direct supervision of medical authorities.
- User Roles: Intended for use by Verified Clinicians and Medical Residents.
- Regulatory (HIPAA/GDPR): For real-world patient data, SightX requires a Supabase B2B/Enterprise Plan.
- Isolated Infrastructure: Each institution must host a separate, sovereign database instance.
- Encryption: Enterprise-grade encryption at rest and in transit is mandatory for HPI (Health Protected Information).
Important
Clinical Disclaimer: SightX is currently a research and educational project. It is intended to demonstrate the potential of AI in telemedicine and should not be used as a primary diagnostic tool without clinical validation and institutional approval.
Built with care to serve people