Skip to content

Repository files navigation

TrustOS

The AI Integrity Layer for a World Powered by AI

Stars Forks Chrome MV3 FastAPI Backend Python Status License Security

We didn't build it to win. We built it because it needed to exist.

Aarav    Priyanshu    Parth    Argha


Can We Trust the Result?

TrustOS provides explainable AI integrity verification for interviews, coding assessments, online examinations, and remote work.


About

TrustOS is an AI Integrity Platform that helps organizations verify whether a candidate genuinely completed an interview, assessment, coding challenge, certification, or remote work session.

Instead of relying on plagiarism detectors or simple AI classifiers, TrustOS combines multiple behavioral signals into an Explainable Trust Score backed by real evidence.


Repository Structure

Hackathon-TrustOS/
├── Chrome_Extension/          Chrome MV3 monitoring agent
│   ├── src/
│   │   ├── background/        Service worker + state management
│   │   ├── content/           Content script for page signals
│   │   ├── offscreen/         Media recording (audio + screen)
│   │   ├── popup/             React popup UI (framer-motion)
│   │   ├── detectors/         Tab, AI, behaviour, screenshot monitors
│   │   ├── services/          API client, WebSocket, session helpers
│   │   └── storage/           Event queue, Chrome storage wrapper
│   ├── public/icons/          Extension icons
│   ├── package.json
│   └── README.md
│
├── Windows_App/               Python desktop tracking application
│   └── README.md
│
├── Presentation/              Project presentation
│   └── TrustOS_Presentation.pdf
│
├── .gitignore
├── README.md
└── run-all.bat

Chrome Extension

The candidate-side monitoring agent that captures authorized session signals.

Feature How
Screenshots chrome.tabs.captureVisibleTab (every 20s)
Screen recording MediaRecorder in offscreen document (30s chunks)
Microphone getUserMedia + MediaRecorder in offscreen document
Speech transcripts Web Speech API via voice popup window
Tab/URL tracking chrome.tabs listeners + content script
AI tool detection URL matching against known AI domains
Clipboard Paste/copy listeners (character counts only, no content)
Idle detection chrome.idle API
Event queue Chrome storage.local FIFO (cap 2000) + batch flush

Key details:

  • Speech recognition runs in a small popup window (not offscreen doc — Chromium limitation)
  • Events flushed to backend every 60 seconds via POST /api/events
  • Per-session config fetched from GET /api/extension/session-config/{id}

See Chrome_Extension/README.md for full architecture.


Backend API

FastAPI async backend with SQLite (default) or PostgreSQL.

Endpoint Purpose
POST /api/auth/login JWT authentication
POST /api/session/create Create monitoring session
POST /api/events Batch event ingestion
POST /api/evidence/upload Upload screenshots, audio, video, transcripts
POST /api/analyze/session/{id} Run AI trust analysis
GET /api/report/{id}/pdf Download PDF trust report
GET /ws/session/{id} WebSocket live event feed
GET /health Health check

AI Engine: Mistral API for behaviour analysis, plagiarism detection (5-gram overlap), and report generation. Runs in offline mock mode when no API key is set.

Scoring model:

  • Browser events (30%): tab switches, URL patterns, clipboard
  • AI usage (30%): frequency, duration, tool type
  • Behaviour (20%): keystroke patterns, idle time, typing speed
  • Document (20%): reference document comparison

Windows App

Python-based behavioural analysis engine that tracks face and hand movement using computer vision.

Feature Purpose
Head pose estimation Detects yaw/pitch/roll — looking away from screen
Gaze direction Tracks where candidate is looking
Hand-gesture detection Phone usage, whispering, notes checking
Face presence Detects face absent, multiple faces
Hand near face Potential phone or note usage
Movement intensity Overall body movement level
Camera access Real-time webcam via OpenCV + MediaPipe
System tray Runs silently with start/stop control

Note: Screenshots, tab tracking, and clipboard monitoring are handled by the Chrome Extension, not the Windows App.

See Windows_App/README.md for full details.


Tech Stack

Layer Technology
Chrome Extension Manifest V3, TypeScript, React 18, Tailwind CSS, framer-motion
Build Vite 6
Backend FastAPI (async), SQLAlchemy, Alembic
AI Mistral API (large-latest, embed)
Database SQLite (default), PostgreSQL
Auth JWT (HS256), bcrypt
Desktop Python 3.11+, OpenCV, MediaPipe, pystray
PDF Reports ReportLab
Communication REST + WebSockets

Quick Start

Backend

cd Chrome_Extension/backend
python -m venv .venv
.\.venv\Scripts\activate
pip install -r requirements-dev.txt
Copy-Item .env.example .env
alembic upgrade head
python -m uvicorn app.main:app --reload --port 8000

Dashboard (Proctor UI)

cd Chrome_Extension/dashboard
npm install
npm run dev
# http://localhost:5174

Extension

cd Chrome_Extension/extension
npm install
npm run build
# Load dist/ in chrome://extensions

One-click (all servers)

Chrome_Extension\run-all.bat

Login: proctor@trustos.dev / proctor123


Presentation

The complete project presentation is available in the Presentation/ folder.

View TrustOS Presentation (PDF)

Click the link above — GitHub renders a built-in PDF viewer when you open the file.

The presentation covers:

  • Problem Statement
  • Product Vision
  • System Architecture
  • Workflow
  • Technical Stack
  • Demo Walkthrough
  • Future Roadmap

Security

  • JWT authentication with token refresh
  • Privacy-first design — clipboard contents never captured, keystroke counts only
  • Evidence encrypted before upload
  • Rate limiting on all endpoints
  • Audit logging for all state changes
  • CORS restricted to known origins
  • .env files excluded from version control
  • bcrypt password hashing
  • MIME type validation on evidence upload
  • Session-scoped data isolation

Read more:


Vision

TrustOS aims to become the integrity layer between Humans, AI, and Digital Evaluations.

The goal is not to prohibit AI, but to provide transparent evidence explaining how AI was used and whether the outcome can be trusted.


Roadmap

Phase 1 — Chrome Extension, Backend APIs, Event Collection, Session Tracking

Phase 2 — AI Behaviour Engine, Explainable Trust Score, Report Generation

Phase 3 — Enterprise Dashboard, Multi-Organization Support, Analytics, API Integrations


Contributors

Built by:

Name GitHub
Aarav Goel @coderaarav12
Priyanshu Mishra @pm5120-alt
Parth Vats @parthvats-code
Argha Atta @TrueIncident

License

This project is licensed under the MIT License.


Building Trust in the AI Era

About

TrustOS is an AI Integrity Platform that verifies whether a candidate genuinely completed an interview, assessment, coding challenge, or remote work session. Instead of relying on simple AI classifiers, TrustOS combines browser signals, face/hand behavioural tracking, and multi-agent AI analysis into an Explainable Score backed by real evidence

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages