Pariksha AI

AI-proctored examination platform with biometric authentication, real-time violation logging, and multi-format assessments.

Features

• Face-authenticated onboarding for both students and examiners.
• Real-time AI proctoring (YOLOv8) that ingests 6‑second webcam chunks, stores violations in MongoDB, and adjusts trust scores.
• Flexible question bank supporting MCQ, subjective, and coding items with reusable Question documents.
• Targeted test delivery via allowedStudents (email lists) and open tests (empty list).
• Examiner tooling for dashboard stats, manual test creation, AI-generated question drafts, and seeding utilities.

Tech Stack

• Frontend: Vite + React + TypeScript + TailwindCSS/ShadCN.
• Backend: Node.js, Express, TypeScript, MongoDB/Mongoose.
• Proctoring ML: Python (YOLOv8 via Ultralytics) invoked per chunk.

---

Prerequisites

Dependency	Version	Notes
Node.js	≥ 18	Needed for both client & server
npm	≥ 9	Ships with Node 18+
Python	3.10–3.12	For the proctoring model (virtual env recommended)
MongoDB	6.x+	Local or Atlas cluster
FFmpeg	Latest	Required to extract frames from WebM chunks

Installing FFmpeg

Windows (Chocolatey):

choco install ffmpeg

Or download from ffmpeg.org

macOS:

brew install ffmpeg

Linux (Debian/Ubuntu):

sudo apt-get update && sudo apt-get install ffmpeg

Verify installation:

ffmpeg -version

---

Project Setup

git clone https://github.com/SanchitNegi177/Pariksha_AI.git
cd Pariksha_AI

1. Install Node dependencies

cd server
npm install

cd ../client
npm install

Important: The backend now uses the CPU-only TensorFlow JS backend for Windows, so it avoids native @tensorflow/tfjs-node compilation and does not require Visual Studio C++ build tools.

2. Prepare the Python virtual environment

The ML proctoring model requires Python 3.10+ with specific packages. Choose one method:

Option A: Using Conda (Recommended)

# Create conda environment
conda create -n venv python=3.10
conda activate venv

# Install dependencies
pip install -r server/src/ml/requirements.txt

# Get the Python path (save this for .env)
python -c "import sys; print(sys.executable)"

Option B: Using venv

cd server/src/ml
python -m venv venv

# Windows
venv\Scripts\activate
# macOS/Linux
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Get the Python path (save this for .env)
python -c "import sys; print(sys.executable)"

3. Configure environment variables

Create server/.env (see template below) and client/.env.

server/.env

PORT=3000
NODE_ENV=development
CORS_ORIGIN=http://localhost:5173
MONGODB_URI=mongodb://localhost:27017/pariksha_ai
OPENAI_API_KEY=
GEMINI_API_KEY=
ANTHROPIC_API_KEY=
JWT_SECRET=
JWT_EXPIRES_IN=
ADMIN_ACCESS_KEY=
ATTEMPT_MAX_ALLOWED_EXITS=

# CRITICAL: Point to your Python environment with ML dependencies installed
# Windows (Conda):
# PROCTORING_PYTHON=C:\Users\your-username\anaconda3\envs\pariksha-ai\python.exe
# Windows (venv):
# PROCTORING_PYTHON=C:\full\path\to\Pariksha_AI\server\src\ml\venv\python.exe
# macOS/Linux (venv):
# PROCTORING_PYTHON=/full/path/to/server/src/ml/venv/bin/python

# If not set, Node.js will auto-detect from CONDA_PREFIX or PYTHON_PATH env vars
PROCTORING_PYTHON=

Important: Set PROCTORING_PYTHON to the absolute path of the Python interpreter where you installed ML dependencies. Use the path from python -c "import sys; print(sys.executable)" above.

client/.env

VITE_API_BASE_URL=http://localhost:3000
VITE_PORT=5173
VITE_NODE_ENV=development

4. Optional: Seed data

After the server env is configured:

cd server
npx ts-node scripts/seedSampleTest.ts

This creates a sample “AI Fundamentals Practice Test” assigned to sanchitnegi177@gmail.com.

---

Running the App

# Backend
cd server
npm run dev

# Frontend (new terminal)
cd client
npm run dev

The Vite dev server prints the local URL (default http://localhost:5173).

---

Using the Platform

1. Register an examiner and/or student (face capture required).
2. Create a test manually or via AI generation (supports MCQ, coding, subjective).
3. Assign students by email using the allowedStudents field (leave empty to make the test visible to all authenticated students).
4. Students log in, review rules, and start their test. Webcam/audio streams are chunked every 6 seconds for proctoring.
5. Violations (multiple faces, devices, missing student, etc.) are logged to ProctoringLog, snapshots stored via GridFS, and ExamAttempt.trustScore is reduced automatically.
6. Submissions store responses, scoring, and trust summaries; no bulky recordings are persisted.

---

Verifying Proctoring

1. Keep the backend terminal visible; you’ll see entries such as:

   ⚠ Violation detected for student 65f... : Multiple faces detected (high)
   

2. Inspect MongoDB:

   use pariksha_ai
   db.proctoringlogs.find().sort({createdAt:-1}).limit(5)
   db.examattempts.find({}, {testId:1, studentId:1, trustScore:1, totalViolations:1})

3. Ensure FFmpeg is installed and the PROCTORING_PYTHON path is valid; errors print directly in the backend log.

---

Troubleshooting

• "Missing required fields" on submit: restart the backend to load the latest server code (submission no longer uploads blobs).
• No violations despite cheating:
  • Verify Python environment has ultralytics opencv-python numpy torch torchvision installed
  • Confirm FFmpeg is installed: ffmpeg -version
  • Check backend logs for "[ML]" and "[PROCTOR]" prefixes
  • Ensure video chunks are being received every 6 seconds
• Client can't reach API: confirm VITE_API_BASE_URL matches the running server and that CORS_ORIGIN allows the client URL.
• Python binary not found:
  • Check that PROCTORING_PYTHON env var points to correct Python executable
  • Verify the path exists: Test-Path "C:\path\to\python.exe" (Windows) or test -f /path/to/python (macOS/Linux)
  • If using conda, ensure environment is activated when setting PROCTORING_PYTHON
• WebM video won't process:
  • FFmpeg must be installed and accessible from PATH
  • Check FFmpeg installation: ffmpeg -version
  • On Windows, you may need to restart terminal after installing FFmpeg for PATH to update

---

Useful Commands

Action	Command
Start backend	cd server ; npm run dev;
Start frontend	cd client; npm run dev;
Build backend	cd server; npm run build;
Build frontend	cd client; npm run build;
Seed sample test	cd server; npx ts-node scripts/seedSampleTest.ts;

---

📄 More Details

See full documentation here → INFO.md

License

No license has been applied to this project yet.