Spooky Machine: A Local Qwen Ghost Story Generator

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine

What I Built

Spooky Machine is a small ghost story generator. Give it a premise—like a locked bedroom door opening at midnight—and it asks a locally running open-weight model to turn that idea into an atmospheric story.

I started with the challenge’s smart-home-sensor prompt, then pared the project back to one thing: telling scary ghost stories. No sensors, accounts, saved stories, or extra controls—just an idea and a story.

Demo

The MVP runs locally. To try it on a computer with Ollama and Qwen3 4B installed:

ollama create spooky-qwen3 -f Modelfile
python server.py

Then open http://127.0.0.1:8000, optionally enter a story idea, and click Tell me a ghost story.

Code

Junie06 / spooky_machine

Spooky Machine

A tiny ghost story generator powered by a local open-weight model.

Run

Install Python 3.11+ and Ollama, then run:

ollama create spooky-qwen3 -f Modelfile
python server.py

The Modelfile makes a small spooky-qwen3 variant from your already-installed Qwen3 model and disables its reasoning prefix, so the app gets the story instead of the model's planning text. Open http://127.0.0.1:8000, optionally add a story idea, and ask for a ghost story. Your idea stays on this computer.

If you don't have qwen3:4b yet, run ollama pull qwen3:4b first. Run the tests with:

python -m unittest discover -s tests -v

View on GitHub

How I Built It

The interface is plain HTML, CSS, and JavaScript. A small Python standard-library server sends the prompt to Ollama’s local generation endpoint and returns the finished story. Qwen3 4B provides the writing; the custom Ollama Modelfile keeps the setup reproducible. The server removes a completed Qwen <think> block before showing the story and refuses an incomplete response rather than displaying planning text.

The app accepts a short, optional idea, limits its length, and renders the model response as text. It has no database or cloud AI dependency. The project is released under the MIT license.

Why Open Innovation Matters

A ghost-story prompt is personal and creative. Running Qwen locally means the idea can stay on my own computer instead of being sent to a hosted AI API. An open-weight model and a small, readable app also make the experience easier to inspect, adapt, and keep running without depending on a particular writing service.

For this MVP, local inference is not an optional add-on: it is the thing that makes the generator private and self-contained.

The Spirit of It

The first version asked the house what happened while everyone slept. The smaller version asks a simpler question: what if the thing in the dark could tell its own story?