ProctorIQ — AI-Powered Secure Assessment & Integrity Platform

Built for PS-003: AI-Powered Secure Assessment & Integrity Platform (EdTech track).

A working prototype of an end-to-end assessment lifecycle: question delivery, in-browser AI proctoring, a sandboxed code judge, plagiarism detection, and explainable risk scoring for recruiters — not just a score, but a breakdown of how the system got there.

Design system

The UI runs on the "Professional Integrity" design language (Plus Jakarta Sans, warm cream surfaces, deep navy text, a terracotta call-to-action accent, and a sage secondary tone). Every page shares one set of CSS variables and component classes in client/src/styles.css, so the palette, spacing, and shape language stay consistent from the sign-in screen through the recruiter analytics dashboards. The layout is responsive throughout — the top navigation collapses to a hamburger menu, stat grids and two-column panels stack on narrow screens, and data tables scroll horizontally instead of overflowing.

Quick start (runs entirely on your laptop, no cloud services needed)

Requires: Node.js 18+ and Python 3 (used only to execute candidates' Python submissions — the app itself is all Node/React).

npm run install:all   # installs server + client dependencies
npm run dev            # starts API on :4000 and the app on :5173

Open http://localhost:5173. Sign in with email + password (both are required and verified against a bcrypt hash), or create a brand-new account from the Sign Up link. The login page also lists the seeded demo credentials:

Account	Email	Password	Role
Candidate 1	student@example.com	student123	Candidate
Candidate 2	student2@example.com	student123	Candidate
Recruiter	recruiter@example.com	recruiter123	Recruiter

Log in as a candidate to take the sample exam. Log in as the recruiter to see the dashboard, per-candidate risk reports, and plagiarism matches (try submitting near-identical code for two different candidates on the same exam — the similarity engine will flag it).

What's implemented

1. Assessment lifecycle MCQ + coding questions, timed attempts, question navigation, autosave of answers as you go.

2. Secure code judge Server-side sandboxed execution (Python & JavaScript) against visible sample tests and hidden tests revealed only at submission — like a real online judge, not a toy checker. server/lib/codeRunner.js

3. Multimodal AI proctoring Runs entirely in the candidate's browser via face-api.js (TinyFaceDetector):

• Live webcam feed with face-presence detection
• Multiple-face detection (flags a possible second person)
• Tab-switch / window-blur detection
• Fullscreen-exit detection
• Copy-paste-into-editor detection

Every event is timestamped and logged to the backend as it happens — nothing is inferred after the fact. client/src/components/ProctorWebcam.jsx

4. Plagiarism detection A MOSS-style k-gram winnowing fingerprint algorithm compares each candidate's code against every other candidate's submission for the same question and reports a similarity percentage plus who matched whom. server/lib/plagiarism.js

5. Explainable risk scoring A transparent, rule-based scorer — every point on the 0–100 scale traces back to a named, weighted factor (face missing, multiple faces, tab switches, paste events, plagiarism %) shown as a bar chart in the recruiter report. No black box. server/lib/riskEngine.js

6. Recruiter analytics Dashboard of all attempts per exam, sortable by score, with a one-click drill-down into each candidate's full report: score breakdown, proctoring timeline, plagiarism matches, and their actual submitted code.

Architecture

secure-assess/
├── server/              Express API + JSON file datastore (lowdb — zero setup, no DB install)
│   ├── index.js         All REST routes
│   ├── lib/
│   │   ├── codeRunner.js    sandboxed Python/JS execution + test grading
│   │   ├── plagiarism.js    winnowing fingerprint similarity
│   │   └── riskEngine.js    explainable weighted risk scoring
│   └── db.js             seed data: 1 exam, 2 MCQs, 2 coding questions
└── client/              React (Vite) — candidate exam UI + recruiter dashboard
    ├── src/pages/         Login, ExamList, CandidateExam, ExamResult,
    │                      RecruiterDashboard, ExamAttempts, CandidateReport
    └── src/components/    ProctorWebcam (face-api.js integration)

No external services, API keys, or paid infrastructure required — everything (including the "AI" proctoring model) runs locally, which also means candidate video never leaves their browser.