TruthEye is an online exam proctoring platform for universities. It combines real-time browser monitoring, AI-assisted face and object detection, and a risk-scoring engine that helps exam supervisors review suspicious sessions. It also covers the academic side of a faculty: courses, doctors, students, proctors, and staff roles.
- Live exam monitoring: students stay connected through heartbeats, and proctors follow sessions in real time over SignalR.
- AI-assisted detection: MediaPipe runs in the browser to detect faces, and a Python SBERT service supports answer grading.
- Integrity events and risk scoring: events such as tab switches, fullscreen exits, copy/paste, multiple faces, and duplicate logins are weighted into a per-session risk score (low, medium, or high).
- Session review: reviewers get a sorted, filterable list of sessions with the evidence attached to each one.
- Role-based access: SuperAdmin, Dean, Faculty Secretary, Technical Staff, Doctor, Student, Proctor, and Support, each with its own permissions.
- Secure account setup: new accounts receive a one-time setup link by email. Administrators never see or send passwords.
- Bulk import: students can be imported from Excel (
.xlsx) or CSV files. - Reports: PDF reports (QuestPDF) and Excel exports (ClosedXML), with protection against spreadsheet formula injection.
- Messaging and notifications: real-time chat and notifications over SignalR.
| Layer | Technology |
|---|---|
| Backend API | ASP.NET Core 9, Entity Framework Core 9, SQL Server |
| Real-time | SignalR hubs (chat, notifications, exam monitoring) |
| Authentication | JWT bearer tokens, rotating refresh tokens in HttpOnly cookies, per-request session validation |
| Frontend | Next.js 16 (App Router), React, TanStack Query, react-hook-form, zod |
| Browser AI | MediaPipe (self-hosted models) |
| AI grading service | Python, Sentence-BERT (SBERT), Docker |
| Logging | Serilog |
| Tests | xUnit with an in-memory database |
TruthEye/ ASP.NET Core API (controllers, services, EF Core, migrations) TruthEye.Tests/ Backend tests Front_End/truth-eye-web/ Next.js frontend TruthEye/AI-BERT-Service/ Python SBERT grading service (Docker) .github/ CI, CodeQL, and Dependabot configuration
- .NET 9 SDK
- Node.js 22 or later
- SQL Server (local instance or LocalDB)
- Docker (optional, only for the SBERT service)
Secrets are never stored in the repository. Set them with user-secrets from the TruthEye folder:
cd TruthEye
dotnet user-secrets set "ConnectionStrings:DefaultConnection" "<your SQL Server connection string>"
dotnet user-secrets set "Jwt:Key" "<a random key of at least 32 characters>"
dotnet user-secrets set "Smtp:Password" "<your SMTP app password>"Other settings, such as the SMTP host and the frontend reset URL, are in appsettings.json.
dotnet ef database updateSet Setup:AllowSuperAdminRegistration to true temporarily, register the SuperAdmin account, then set it back to false.
dotnet run --project TruthEyeCreate Front_End/truth-eye-web/.env.local:
NEXT_PUBLIC_API_URL=http://localhost:5210
Then run:
cd Front_End/truth-eye-web
npm ci
npm run devcd TruthEye/AI-BERT-Service
docker compose up --buildcd TruthEye.Tests
dotnet test- Secrets (connection strings, JWT keys, SMTP passwords) must be set with user-secrets or environment variables, and must never be committed.
- The API exposes a Content Security Policy report endpoint at
/api/csp-report, and the frontend uses per-request nonces. - Dependencies are monitored with Dependabot, CodeQL scans run weekly, and CI blocks packages with known vulnerabilities.
- To report a vulnerability, see SECURITY.md.
- Exam presence and the per-session closing lock live in process memory, so the API must run as a single instance. Moving to several instances needs a shared store for both.
- Profile photos are served from
wwwroot/uploads/profileswith random file names. Anyone holding the link can open the photo. - Deleting a dean, secretary or support account only deactivates it, so their history stays intact.
No license has been chosen yet. All rights are reserved by the author until one is added.