LeafLock — Open-Source AI Outdoor Steganography Engine (#hf26challenge)

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

What I Built

LeafLock is an asynchronous, two-player botanical steganography web app that turns nature into a cryptographic key vault. Instead of digital passwords, users lock secret messages behind physical, biological objects in the real world. Player 1 encrypts a message behind a plant, and Player 2 must step outside, solve an AI-generated poetic riddle, and scan the correct physical object with their phone camera to unlock it.

Demo

• Live Demo: https://leaflock.onrender.com/

(Drag and drop your outdoor field-test photo here!)

Code

iqbhavya / LeafLock

🍃 LeafLock — Open-Source AI Outdoor Steganography Engine

Built for Hacktoberfest 2026 (#hf26challenge), Global Hack Week (GHW), and DEV.to Hackathon Challenges.

LeafLock applies digital steganography to the physical world. Player 1 locks a secret payload message behind any physical outdoor object (nature, animals, plants, items). Player 2 must decode an AI-generated riddle, step outdoors, find the object, and capture a matching photo with their camera to decrypt the message.

---

🏆 Hacktoberfest Prize Categories & Alignment

• Touch Grass Challenge (#hf26challenge): Asynchronous 2-player outdoor game that encourages participants to step outside, explore flora/fauna, and engage with nature.
• Best Use of Gemma ($200): Uses Google AI Studio and Google Gemma models (gemma-4-31b-it, gemma-4-26b-a4b-it) to dynamically synthesize 2-line cryptic riddles based on real-time detected botanical, object, and color features.
• Best Use of MongoDB Atlas: Uses MongoDB Atlas (M0 Cluster) with Mongoose models…

View on GitHub

How I Built It

LeafLock relies on a hybrid edge-and-cloud architecture to keep processing fast and privacy-centric:

• Frontend: The user interface was built using React 19 and Vite 8, styled with Tailwind CSS, and utilizes HTML5 environment camera capture.
• Backend & Database: A Node.js/Express backend connects to a MongoDB Atlas M0 cluster to manage cipher state, async multiplayer sessions, and completion tracking without permanently storing raw image files.
• Open-Source Vision Pipeline: The image analysis integrates @tensorflow-models/mobilenet for object classification alongside sharp to extract perceptual dHash fingerprints and 64-bin color histograms.
• Dynamic Riddle Generation: The backend queries Google's Gemma 2 model via the Google AI Studio API. It prompts Gemma 2 to dynamically synthesize 2-line cryptic riddles based on the botanical features detected by the vision pipeline.
• Deployment: The React/Vite frontend asset pipeline is hosted and deployed on Render's global edge network.

Why Does Open Innovation Matter?

When users explore nature, their surroundings shouldn't feed closed corporate surveillance models. By relying on an open-weight ecosystem—Google's Gemma 2 and local-first vision processing—LeafLock ensures that raw photos are evaluated transparently and discarded instantly. Open innovation makes it possible to build location-free, privacy-preserving outdoor games that can eventually run entirely on edge hardware deep in a national park with zero cellular reception.

My Agent Session

I scaffolded and iterated on this architecture using Antigravity paired with the DevRelay MCP server, tracking challenge telemetry and evaluating test prompts locally.

LeafLock — Open-Source AI Outdoor Steganography Engine Session

You

Building LeafLock: Open-source botanical steganography web app for Hacktoberfest Touch Grass Challenge powered by Google AI Studio Gemma models, MobileNet AI vision, and MongoDB Atlas.

Agent

Configured full-stack React + Node Express application with Google AI Studio Gemma model integration, MobileNet vision classifier, sharp dHash perceptual fingerprinting, and MongoDB Atlas database.

2 of 2 messages

Prize Categories

I am entering LeafLock into the following Hacktoberfest prize tracks:

• Overall Winner (\$250): Built an original, sensor-driven outdoor gameplay mechanic that forces users to disconnect from their monitors and engage with physical botany.
• Best Use of Gemma (\$200): Leveraged Google's open-weight Gemma models (via Google AI Studio) to dynamically craft atmospheric, cryptic outdoor riddles from visual labels without spoiling target names.
• Best Use of MongoDB Atlas (\$100): Used a MongoDB Atlas M0 Cluster with Mongoose to securely manage shortCodes, extracted visual features, riddles, and completion states.
• Best Use of Render (\$200): Deployed the React/Vite web application and frontend asset pipeline on Render's global edge network to ensure fast, highly available access for users exploring outdoors.