DEV Community: NAITIK AGARWAL

WEEK 3 CHALLENGE

NAITIK AGARWAL — Fri, 23 Oct 2026 10:27:04 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

What I Built

Ghostwriter for Reality — From scattered fragments to finished work.

Ever abandoned a project and returned to it weeks later, only to realize you've forgotten why you made certain decisions, what remained unfinished, or what you were supposed to build next?

That's the problem I'm trying to solve.

Ghostwriter for Reality helps developers, creators, and makers recover the context behind stalled projects. Instead of starting from scratch, they can bring together scattered notes and project fragments, reconstruct their original intent, identify uncertainties, and turn everything into an actionable recovery plan.

For this first version, I'm focusing on a Markdown-first workflow. The goal is to prove the core experience before expanding into more complex document types and richer project recovery tools.

Less time reconstructing your own thoughts. More time actually building.

Demo

🚀 Live App: https://ghost-writer-vert.vercel.app/

Code

💻 GitHub Repository: https://github.com/Naitik-xd/Ghost-Writer-

The project is open source, and contributions and feedback are welcome!

How I Built It

I'm building Ghostwriter for Reality around Gemma 4, with the aim of using open-weight AI for project reconstruction and writing assistance.

The application is built as a web-based workspace, with a focus on a clean interface, local project persistence, and a workflow that keeps the user in control of their original content.

The first version focuses on the essentials:

Organizing project fragments.
Reconstructing project intent and identifying uncertainties.
Turning recovered context into actionable next steps.
Generating useful project documentation from scattered information.

I'm taking a Markdown-first approach for this initial release. This isn't about reducing the ambition of the idea; it's about building a reliable foundation that can grow into the complete vision.

Why Does Open Innovation Matter?

A tool that helps people recover their work should give them control over that work, too.

Open innovation makes it possible for developers to inspect the implementation, suggest improvements, adapt the workflow, and build on the project rather than being limited to a closed product.

Open-weight AI also creates opportunities to experiment with model behavior, build transparent workflows, and develop alternatives that the community can improve together.

Ghostwriter for Reality is an attempt to make that practical: help people recover unfinished ideas and turn them into something they can continue building.

Slow and steady isn't about thinking smaller. It's about building something useful, one reliable step at a time. 🚀

Submaster-ai Dev Challenge week 2

NAITIK AGARWAL — Fri, 16 Oct 2026 14:47:50 +0000

This is my submission for Hacktoberfest Open-Source AI Challenge — Week 2: Side Hustle.

🚀 What I Built

SubMaster AI — An AI-powered teacher substitution and staffing management system for K–12 schools.

When teachers are absent, administrators often spend hours manually rearranging timetables. SubMaster AI automates this process in under 5 seconds, preventing scheduling conflicts, balancing workloads, and generating substitute lesson briefings.

Business Model: A proposed subscription of $250/month for private and charter schools.

🔗 Demo & Source Code

Live Demo: https://submaster-ai.vercel.app/
GitHub: https://github.com/Naitik-xd/Dev-3

🛠️ Tech Stack & Architecture

AI: Google's open-weight Gemma models via @google/genai
Frontend: React 19, TypeScript, Tailwind CSS
PDF Export: jsPDF
Backend: Vercel serverless functions

A deterministic TypeScript constraint solver handles scheduling and workload rules, while Gemma generates contextual lesson briefings and emergency planning suggestions.

🌍 Why Open Innovation Matters

Privacy: Open-weight models offer opportunities for auditable and locally hosted AI solutions.
Affordability: Schools can explore deployment options that reduce reliance on proprietary AI APIs.
Adaptability: Open foundations make it easier to customize scheduling rules for different schools.

SubMaster AI combines deterministic scheduling with AI assistance to make school staffing faster, smarter, and more reliable.

Touch Grass, Week 1 Dev Challenge hacktoberfest 26

NAITIK AGARWAL — Wed, 07 Oct 2026 10:42:46 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

TouchGrass is a web app that uses AI to turn screen time into a reason to go outside.

Instead of endlessly scrolling, you get a small outdoor challenge — like finding a natural pattern, capturing a leaf, or observing something in the sky. You complete it using your camera, and AI checks whether your outdoor evidence matches the challenge.

The app tracks both your current session time and the time spent completing the challenge, encouraging you to actually step away from the screen and interact with the world around you.

It’s designed for anyone who wants a simple push to get outside without turning it into another social or productivity app.

Demo

Live Demo:

touchgrassai.vercel.app

Code

GitHub:

Naitik-xd / Week-1-Dev-Challenge

Hacktoberfest 26

TouchGrass

An anti-screen habit application built for Hacktoberfest 2026 Week 1 ("Touch Grass"). TouchGrass assigns micro outdoor observation challenges, encourages you to physically step away from your device into nature, and verifies your findings through live camera evidence.

AI Architecture

TouchGrass is powered strictly and exclusively by Google's open-weight Gemma 4 models via the @google/genai SDK:

Primary Model: gemma-4-26b-a4b-it
Fallback Model: gemma-4-31b-it

Functions Handled by Gemma 4:

Dynamic Challenge Generation (POST /api/challenge): Synthesizes safe, achievable outdoor observation tasks across 13+ varied themes with randomized seeds to prevent repetition.
Multimodal Vision Verification (POST /api/validate): Analyzes camera-captured snapshots against the active assignment criteria and delivers constructive feedback.

Core Features

Interactive 3D Terra Biosphere: Three.js WebGL centerpiece on the hero page with cursor-reactive rotation and floating atmospheric particles.
Strictly Camera-Only Shutter: Evidence capture invokes the device hardware photo camera directly (capture="environment"…

View on GitHub

How I Built It

TouchGrass is built with React, TypeScript, Tailwind CSS, Vercel, and Gemma.

Gemma is used to generate outdoor challenges and analyze camera-captured evidence to determine whether it reasonably matches the assigned task.

The app uses a server-side API route to keep the AI API key out of the client, while session and timer state are persisted locally so refreshing the page doesn't reset an active challenge.

There’s no login, database, or social feed — the focus is simply:

Get a challenge → Go outside → Capture evidence → Complete it.

## Why Does Open Innovation Matter?

Open innovation makes it possible to build experiences where AI is not just another chatbot sitting inside a screen.

With open-weight AI models like Gemma, developers can experiment with AI as part of real-world interactions — in this case, using AI to encourage people to leave the screen and engage with their physical environment.

That flexibility makes it easier to build small, focused ideas around AI without needing to turn everything into another closed, screen-first experience.

## Prize Categories

Gemma

CurriculumCraftai, Submission for launch weekend DEV Challenge

NAITIK AGARWAL — Sat, 03 Oct 2026 06:35:37 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

## What I Built 🎓✨
I built CurriculumCraft AI for my friend who teaches high school (CBSE Grades 9–12) in India.

Teachers spend hours every weekend formatting exam papers and calculating question weightages. CurriculumCraft AI automates this in seconds by generating:

📄 Student Question Papers (MCQs, Assertion-Reason, Short Answer, Case Studies) aligned with rationalized NCERT syllabi.
📝 Confidential Teacher Marking Schemes with step-by-step scoring rubrics.
🗓️ NEP 2020 Monthly Lesson Plans with period-by-period trackers.
🖨️ Print-Ready A4 & PDF Exports with KaTeX math and science equation formatting.

**## Demo 🌐

Live App:**

curriculum-craft-ai.vercel.app

Code 💻

https://github.com/Naitik-xd/CurriculumCraft-AI

License: MIT License
Author: Naitik

## How I Built It 🛠️🤖

- Open-Source AI: Powered strictly by Google's open-weight Gemma 4 family (gemma-4-26b-a4b-it & gemma-4-31b-it).
- Hallucinations: Low temperature (0.2) and pedagogical blueprint prompts ensure strict factual alignment with NCERT chapters.
- Tech Stack: React 19, TypeScript, Tailwind CSS, KaTeX, and Express serverless functions on Vercel.
- Security & Rate Limiting: Server-side API key isolation with a 30 req/5hr rate limiter to prevent token abuse.

*## Why Does Open Innovation Matter? *🌍💡
Closed commercial APIs are cost-prohibitive for educators and lock pedagogical tools behind paywalls. Open-weight models like Gemma democratize high-quality educational AI, allowing developers to build free, privacy-friendly, and transparent tools that teachers can actually rely on without vendor lock-in.

## Prize Categories 🏆

Build for a Friend 🤝
Open Innovation 🌐