System Topology

Nomevex follows a modular monolith architecture with clearly separated concerns:

• Backend API — Fastify-based HTTP server with WebSocket support
• PostgreSQL Database — Primary data store with row-level security
• Redis Cache — Session state, rate limiting, and pub/sub
• SvelteKit Portals — Four separate frontend applications (Admin, Student, Instructor, Verification)
• NAE (NOMEVEX Assessment Environment) — Native Rust application with embedded WebView
• AI Inference Engine — On-device TensorFlow Lite models

Data Flow

During a proctored exam session, data flows through the following pipeline:

1. Client — NAE captures events locally (keystrokes, focus changes, AI detections)
2. Local Processing — AI inference runs on-device; events are hashed into Merkle tree
3. Transmission — Encrypted event metadata sent via WebSocket to backend
4. Backend — Validates, stores, and broadcasts events to authorized viewers
5. Anchoring — Merkle roots periodically anchored to blockchain/transparency ledger
6. Verification — Independent auditors verify session integrity via Verification Portal

Key Design Decisions

• Local-first: students can take exams offline; data syncs automatically on reconnect
• Privacy-preserving: raw video/audio never leaves the client device
• Verifiable: cryptographic proofs enable independent audit without data access
• Scalable: horizontal scaling via stateless API servers and PostgreSQL replication