DEV Community

Cover image for Touch Grass, Not Glass!
Rana Umar
Rana Umar

Posted on

Touch Grass, Not Glass!

#devchallenge #hf26challenge

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

EyesUp — Get Out of the Digital World, Into the Physical World

"Touch grass, not glass."

What I Built

Most modern habit trackers and gamification apps suffer from a hypocritical flaw: they demand you constantly stare at your screen to prove you're being productive.

I built EyesUp — the anti-app designed for radical screen minimization.

Our core motto is simple: "Get users out of the digital world, into the physical world."

EyesUp is not a traditional quest generator locked into rigid, rule-based policies. Instead, it dynamically adapts to your personal habits, provides maximum flexibility, and keeps you immersed in the analog world through open-source AI.

Key Highlights & User Journey

  1. Seamless 4-Step Onboarding:

    In just four quick steps, users configure their profile: set your name, select your favorite hobbies (botany, sketching, walking, cafe exploration, reading), choose your preferred language, and dial in your focus duration (15 to 60 minutes).

  2. Dynamic Quests with Maximum Flexibility (Powered by Gemma):

    Every day, EyesUp serves up fresh, personalized quests tailored to your profile—never repeating yesterday's tasks. And if you aren't feeling the suggested quest? You have complete freedom: simply type your own idea (e.g. "Find a historical lamp post on my street" or "Walk 1,000 steps without checking messages"), and Google's open-source Gemma generates a tailored quest on demand.

  3. Screen-Free Audio Stories with Offline Caching (Piper TTS + Gemma):

    To prevent boredom without forcing you to stare at a phone, EyesUp generates educational Audio Stories about the objects, nature, or locations in your quest. Powered by Gemma and Piper TTS (open-source local neural speech), it narrates fascinating historical lore, evolutionary context, and sensory mindfulness. Crucially, audio stories can be cached for 100% offline use, so you can head into nature without needing active cellular data.

  4. In-Browser Pocket Pedometer:

    Going on a walking quest? EyesUp taps into the browser's native DeviceMotion accelerometer API. Just slip your phone into your pocket—no native app installation or smartwatch needed—and it quietly tracks your footsteps while you keep your eyes up and forward.

  5. AI Vision Verification & Holographic Sticker Rewards:

    After finishing your quest, snap a photo of your physical creation, nature finding, or step summary. Multimodal AI verifies the submission against the quest criteria and awards collectible Holographic Stickers tiered by effort and relevance (Bronze, Gold, Prismatic). You can collect them in your gallery, track weekly streaks, and share achievements with your community.

EyesUp is built for digital workers, students, and anyone burnt out by screen fatigue who wants an encouraging nudge to reconnect with the physical world.

Demo

Dashboard:

Sticket Gallery:

Quest tracking screen:

Credibility Check:

In-Browser Pocket Pedometer:

Code

You can view the full source code on GitHub:

GitHub logo Umar4880 / EyesUp

EyesUp — Radical Screen Minimization

An "anti-app" that rewards you for putting your phone down.

EyesUp generates personalized analog quests (go touch grass, sketch something, walk 1,000 steps) and uses AI vision to verify you actually completed them — earning you streaks, collectible cards, and a healthier relationship with your screen.

Table of Contents

  1. What is EyesUp?
  2. Feature Highlights
  3. Tech Stack
  4. Project Structure
  5. Frontend Navigation
  6. Backend Architecture
  7. Quest Types
  8. Getting Started
  9. Environment Variables
  10. Running Tests

What is EyesUp?

EyesUp is a fullstack web application built as part of the MLH Hackathon. It flips the script on habit-tracking apps: instead of rewarding you for staying on your phone, it sends you out into the real world.

Each day, EyesUp's AI engine (Google Gemini) generates a personalized Quest Plan — a set of short analog…

View on GitHub
  • Frontend: Next.js 16 (App Router), React 19, Tailwind CSS 4, Lucide React, Web Sensors (DeviceMotion API).
  • Backend: FastAPI, Python 3.13, aiosqlite (async SQLite in WAL mode), SQLAlchemy, package-managed with uv.

How I Built It

EyesUp is engineered around open-weight models, local neural speech inference, and multimodal reasoning:

  1. Google Gemma for Structured Quest & Lore Generation:

    We utilize Gemma open-weight models (gemma-4-26b-it) to power dynamic quest generation and deep educational audio scripts. Structured Pydantic schemas enforce clean outputs, guaranteeing rich historical facts and sensory mindfulness prompts for every activity.

  2. Open-Source Neural Speech with Piper TTS:

    Instead of sending private audio or incurring latency from proprietary cloud TTS APIs, EyesUp integrates local ONNX-runtime open neural TTS (piper-tts / Kokoro) directly on the backend, generating .wav streams that can be cached on the device for completely offline outdoor sessions.

  3. Multimodal Vision Verification:

    When verifying proof, the vision engine analyzes physical outdoor context, flora, paper drawings, or pedometer logs. If lighting or angle is imperfect, users get immediate feedback and can retake on the spot without losing their session progress.

  4. Passive Accelerometer Pocket Pedometer:

    A custom React sensor hook taps into devicemotion acceleration vectors with low-pass peak filtering. Users don't need to watch their screen while walking: phone in pocket, head up, eyes forward.

Why Does Open Innovation Matter?

  1. Digital Wellbeing Shouldn't Belong to Ad-Tech Giants: Traditional screen-time apps are built by the very companies profiting from our screen time. Open-source AI allows us to build software whose only incentive is user flourishing.
  2. Privacy on Personal Habit & Sensory Data: Your camera photos, accelerometer movement traces, and voice narration shouldn't be locked into closed corporate silos. Running open-weight vision models and local TTS ensures your personal life stays under your control.
  3. Accessibility & Customization: Open models empower developers everywhere to customize quest generation for regional climates, languages (like Urdu, Spanish, or Hindi), and varying physical abilities without paying per-token gatekeepers.

Prize Categories

  • Grand Prize / Touch Grass Category: Empowering people to disengage from screens and immerse themselves in the physical world through open-source AI.

Built with ❤️ for Hacktoberfest Week 1 by Umar Javed (@Umar4880).

Top comments (0)

Some comments may only be visible to logged-in visitors. Sign in to view all comments.