This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass
What I Built
🌱 Touch Grass is an AI-powered real-world scavenger hunt designed to get people off their screens and exploring the world around them.
The idea is simple: AI generates five random riddles describing real-world objects. Players must find those objects and upload photographs to solve the puzzles.
Each puzzle is worth 1 point, giving players a maximum score of 5 points. They can request hints without losing points or reveal answers early, but revealing an answer earns zero points for that puzzle.
The goal is to turn a walk around campus, a park, or a neighborhood into a fun AI-powered adventure.
Demo
Demo linkL https://youtu.be/65kJiwxRiXQ
Code
🔗 GitHub Repository: https://github.com/aanoop234/Hacktober-fest-week-1-challenge
How I Built It
I'm building Touch Grass using open-weight AI and modern web technologies.
Gemma 3 4B: For generating riddles and exploring image-based answer verification.
Ollama: To run the AI model locally without relying on paid AI APIs.
React + Vite: For a responsive, installable Progressive Web App.
Node.js + Express: For game logic, photo uploads, AI integration, and score management.
Why Does Open Innovation Matter?
Open innovation makes AI experimentation more accessible.
By using an open-weight model with Ollama, this project can explore local AI inference without depending on paid, proprietary AI APIs. It provides more control over how AI requests are processed and makes experimentation possible on ordinary hardware.
For Touch Grass, open innovation means I can explore how AI can be integrated into an everyday experience while learning about prompt engineering, image understanding, structured outputs, and the limitations of local AI models.
Prize Categories
Best Use of Gemma : It is used to generate puzzles and verify your answer
Top comments (1)
tr.ee/dev-to