Don't Wake the Sloth

NOORAS FATIMA ANSARI — Tue, 20 Oct 2026 15:00:41 +0000

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

What I Built

I built Don’t Wake the Sloth, a small puzzle game where you guide monkeys onto banana tiles without making enough noise to wake a sleeping sloth.

The game has ten levels, a noise meter, stars, undo, sound effects, and a calm jungle soundtrack. You can play in the browser or run the terminal version. There’s also a local Gemma assistant called Slowpoke: ask for a hint, or let it choose a move. The game rules—not the model—decide whether that move is legal.

It’s for anyone who likes a puzzle with low stakes, quiet leaves, and a sloth who would really prefer not to be disturbed.

Demo

https://dont-wake-the-sloth.onrender.com
https://dont-wake-the-sloth.vercel.app

Code

https://github.com/nooras/dont-wake-the-sloth

How I Built It

The browser game is built with React and Vite. The game rules are deterministic JavaScript, the ten boards live in levels.json, and a Dijkstra solver calculates the minimum-noise par for each level. There’s also a Python terminal game and agent.

I run Gemma 3 locally with Ollama. Slowpoke can offer hints and commentary, or choose a single move. The game validates every move itself; if Gemma is unavailable or proposes an invalid move in the terminal game, a local solver takes over. No hosted AI API key is needed.

I also made the ambience, lullaby, and action sounds with the browser’s Web Audio API, so the game doesn’t need external audio files.

Why Does Open Innovation Matter?

Running an open-weight model locally let me experiment with an AI character without setting up a paid API or sending game state to a hosted model service. I can change the prompt, try another locally available model, and inspect how the assistant fits into the game.

Most importantly, the AI is helpful without being in charge. The rules stay transparent and testable, while the model adds personality and suggestions.

The Lazy Payoff

No time-saved number measured—I don’t want to invent one. But the solver does the tedious part of finding and checking quiet routes, and gives each level a reproducible par. Players can ask Slowpoke for a nudge instead of restarting just to try a different idea.

Prize Categories

Gemma, Github Copilot, Render

Yogi Buddy: Yoga with me

NOORAS FATIMA ANSARI — Mon, 05 Oct 2026 20:17:33 +0000

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

What I Built

Yogi Buddy: Yoga with Me is a friendly yoga companion that anyone can open and use straight away. There's no login and no sign-up: you open the page, tap start, and a calm chat guide walks you through a session.

I built it for My friends who sit at a desk all day. They wanted to start yoga but felt intimidated by videos and didn't know which poses were safe for a stiff back."

Yogi Buddy asks how you feel, how much time you have, and whether you have any pain or injuries. Then it suggests a routine for goals like stress relief, headache relief, better sleep, back and neck pain, or weight loss, with step-by-step guidance and breathing exercises. It follows safety rules: it isn't a doctor, it asks about pregnancy and injuries first, and it recommends seeing a professional for anything serious.

Demo

Live app: https://yogibuddy.vercel.app/
Backup deployment: https://yogibuddy.onrender.com/ (free tier, so the first load can take about a minute while it wakes up)

Code

https://github.com/nooras/yogibuddy

How I Built It

Open-weight models through Hugging Face Inference Providers. The backend calls Hugging Face's OpenAI-compatible router, using meta-llama/Llama-3.1-8B-Instruct.
Ollama for local development. I started by running models locally with Ollama, then moved to a hosted open-model backend so visitors don't need to install anything.
Backend proxy. The Hugging Face token lives only on the server, and the chat endpoint has rate limiting, since the app has no login.
Swappable model. The model name and token come from environment variables (MODEL_NAME, HF_TOKEN), so I can change models without changing code.
Prompt design. A system prompt defines the friendly personality, the yoga knowledge (styles, poses, breathing, goal-based routines), and the safety rules.
Frontend: [React + Vite], deployed on Vercel. The backend is also deployed on Render.

Why Does Open Innovation Matter?

Open models let me build something personal without being locked into one vendor or one price. A few real moments from this project:

I could start free and local. Running Ollama on my own machine meant I could build and test the whole experience at no cost, which mattered for a weekend project.
I could switch models quickly. While building, a model id I picked didn't exist on the router, and another model returned empty replies. Because the model is just an environment variable, I changed it and redeployed without rewriting anything.
Control over a health-adjacent app. Wellness chats can feel personal. With open models I can choose where it runs, from a hosted API to my own server, and keep the option to move it fully local later.

A closed API would have worked for a demo, but it wouldn't have given me the same freedom to experiment, swap, and self-host.

Prize Categories

Render, Github Copilot

DEV Community: NOORAS FATIMA ANSARI

Don't Wake the Sloth

What I Built

Demo

Code

How I Built It

Why Does Open Innovation Matter?

The Lazy Payoff

Prize Categories

Yogi Buddy: Yoga with me

What I Built

Demo

Code

How I Built It

Why Does Open Innovation Matter?

Prize Categories