2026 · SHIPPED · BUILD

GreenGuardian

Proctored online exams with no media server and no cloud functions

Next.js 16
React 18
TypeScript
Firebase
TensorFlow.js
Gemini AI
Tailwind CSS

GreenGuardian is an AI-powered exam proctoring and academic-management platform. It runs webcam-monitored online exams with real-time proctoring, grades and reviews submissions, catches plagiarism, and manages classrooms, notices, results and live video meetings.

The notable property is what is absent: no media server, no Cloud Functions, and a stack that stays inside a free Firebase plan. Proctoring inference happens in the candidate browser via TensorFlow.js, which means the architecture is cheap and the video is private for the same reason.

Problem

Online exam proctoring normally means a media server, a video pipeline, and a bill. For a university running on a free tier, that architecture is simply unavailable.

Question

How much of a proctoring and academic-management platform can be pushed into the browser and a serverless database before the architecture breaks?

Approach

Move the vision work to the client. BlazeFace and COCO-SSD run in the candidate's own browser through TensorFlow.js, so frames never need to reach a server at all — only the derived signals do. Everything else is Firebase Auth and Firestore, with no Cloud Functions and no media server in the path.

System

Open any stage to see what it does.

BROWSER — The exam runs client-side. Webcam frames stay on the candidate machine.

ON-DEVICE VISION — TensorFlow.js with BlazeFace for face presence and COCO-SSD for object detection. The video never leaves; only structured proctoring signals do.

AUTH + ROLES — Firebase Auth with student, teacher and admin roles gating every workflow.

FIRESTORE — Classrooms, notices, submissions, results. Serverless, and compatible with the Firebase Spark free plan.

AI-ASSISTED MARKING — OCR over submitted work, plagiarism checks, and Gemini-assisted marking — assistance for a human marker, not a replacement for one.

ANALYTICS — Academic analytics over results and participation for teachers and administrators.

Demo

Full student, teacher, admin and classroom workflows: run a webcam-monitored exam, review and grade submissions, check for plagiarism, publish results, post notices, and hold live video meetings.

Result

Runs on a serverless stack with no media server and no Cloud Functions.
Stated by the repository to be compatible with the Firebase Spark (free) plan.
Deploys to Vercel; MIT licensed.
Ships student, teacher, admin and classroom management workflows alongside proctoring, plagiarism detection, OCR and AI-assisted marking.

Lesson

The constraint produced the architecture. Being unable to afford a media server is the only reason the vision runs on-device — and on-device vision turned out to be the better privacy story as well as the cheaper one. I would not have arrived at it with a budget.

Source

github.com/BakulBd/GreenGuardian ↗

Connected

GUB Portal — Both are WEB and CLOUD work, and both are built with Next.js 16, TypeScript.
hdgame.me — Both are WEB and CLOUD work, and both are built with Next.js 16, TypeScript.
TeleFlow — Both are WEB and CLOUD work, and both are built with Next.js 16, TypeScript.
ContextGuard — Connected through COMPUTER VISION and AI.