Inspiration

I wanted one page where a student could paste a draft and the reference sources they used, and immediately see which phrases need a citation — without creating an account, without their writing being sent to a third-party service, and without any API key. That constraint shaped everything: no login, no analytics, no external database, no tracking cookies.

What it does

VidyaSetu compares a draft against pasted reference sources and produces a structured similarity report with matched-phrase evidence chips, per-source match bars and a plain-text report download. It formats citations in APA 7, MLA 9 and IEEE, saves generated references to localStorage and exports them with copy-all. It turns word count and a deadline into a paced assignment schedule with a local checklist, and hosts a sanitized study exchange board for feedback requests. Draft statistics update live, the report can be read aloud, and Ctrl/Cmd+Enter re-runs the check. /privacy.html and /terms.html carry the academic-honesty disclaimer and a one-click local-storage clear control.

How we built it

No build step, no bundler, no framework and zero npm dependencies. The frontend is three local files — public/index.html, public/styles.css, public/app.js — with system fonts only and no CDN or external asset. Five Node.js serverless functions live in api/ (health, check, cite, plan, board) and export CommonJS handlers; server.js runs those same handlers locally with Node's built-in http, fs, path and url modules. vercel.json sets cleanUrls and rewrites /favicon.ico to /favicon.svg. test_suite.mjs uses node:assert/strict and runs with npm test.

The browser side uses fetch for the five endpoints, localStorage for saved references and planner tasks, the Web Speech API for read-aloud, the Clipboard API for copy-all, and Blob + URL.createObjectURL for the report download. The originality engine blends word-level containment with 5-word shingle containment into a per-source match percentage, scores overall as 70% strongest source + 30% average, and merges shared shingles into maximal matched phrases (minimum 6 words) shown as evidence chips.

Challenges I ran into

Keeping the whole backend self-contained: no external API key, no database and no third-party service, while still returning a useful, explainable similarity report. Showing why a phrase was flagged (rather than just a percentage) meant building the evidence-chip merge on top of the shingle comparison.

Accomplishments I'm proud of

Five working modules — Originality Check, Citation Builder (APA/MLA/IEEE), Assignment Planner, Integrity Guide and Study Exchange Board — run on one static frontend plus self-contained serverless handlers. The project has a live public deployment, an automated npm test suite, dedicated privacy/terms pages and a documented API surface.

What I learned

A privacy-first architecture is a feature, not a limitation: zero accounts, zero trackers and zero external API keys still leave room for a complete writing toolkit. Accessibility details and honest self-check limitations matter as much as the algorithm.

What's next for VidyaSetu

More citation styles, additional report export formats, and moderation options for the study exchange board.

Built with

HTML5, CSS3, ES6+ JavaScript, Node.js (>=18), Vercel Serverless Functions, REST APIs, Web Speech API, localStorage, Clipboard API, Blob / Object URL, npm, Git / GitHub, MIT. No React, no Vite, no Tailwind, no external library — package.json has no dependencies or devDependencies.

AI disclosure

The project was developed with AI coding assistance. All significant AI-generated contributions are disclosed here and in AI-USAGE.md; every generated part was read, understood and tested, and I can explain and demonstrate each part of the codebase.

Disclosure on timing

This repository was first created on 27 October 2026, before the IceHacks hacking period. Pre-existing code is disclosed here as required by the official rules. No claim is made that the entire project was written inside the hacking window.

Built With

Share this project:

Updates

Submission history