little tales-touch the grass

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

What I Built

'Little Tales - Touch Grass Edition' for Pre-1 Blue (5-year-olds).

Kids type a word like 'kitty, bunny, leaf, bird, rainbow' for 10 seconds. The app generates a short 5-line story in very simple English and always ends with an OUTDOOR MISSION like:

🌿 OUTDOOR MISSION: Go outside and find 3 different color leaves like Lily!

So the screen time is the shortest part - 10 seconds to generate, 10 minutes outside to actually do the mission. It's built for classrooms where kids need to move, touch grass, smell flowers, hop like bunnies.

Who is it for? Pre-1 to Grade 1 teachers and parents who want stories that get kids OFF the tablet and into the garden.

Demo

https://github.com/eshaal-wq/Little-Tales/tree/main

It works 100% offline after first load. Turn off WiFi and it still generates.

Code

https://github.com/eshaal-wq/Little-Tales/tree/main

Main file is single 'index.html' - no backend, no build. Just open in Chrome.

Fallback offline story DB included for instant use if network is slow.

How I Built It

Open-source AI at core:

• Open-weight model: Llama 3.2 1B Instruct (Meta open weights) + SmolLM2-135M for fast demo - both run locally
• Local inference harness: @mlc-ai/web-llm 0.2.79 - runs WebGPU inference in browser, no server
• No closed API: No OpenAI key, no cloud. First visit caches ∼800MB weights, after that fully offline.

The entire app is built around local inference. Prompt is tuned for Pre-1 vocabulary (<80 words, happy, simple). If I used closed API, it wouldn't work in my school.

Why Does Open Innovation Matter?

Closed AI fails for my classroom in Karachi: no stable WiFi, no budget for API, and I cannot send 5-year-olds' data to a cloud.

Open made this possible:

1. Private & Offline: Kids data never leaves the laptop - critical for schools
2. Free & Accessible: Any teacher can open index.html on an old laptop, no payment
3. Swappable & Customizable: I can swap Llama for SmolLM or Phi-3 for even smaller devices, and change prompts for Urdu/English vocab without asking a vendor
4. Touch Grass by design: Because it runs locally, I can take the laptop to the garden itself.

Open innovation turned AI from a cloud service into a classroom tool that gets kids outside.

My Agent Session

Built with Meta AI assistance for debugging WebLLM loading and optimizing for offline fallback.

Prize Categories

• Touch Grass - Getting people off screen
• For Good / Education