DEV Community: ch Thriveni

"GhostWriter AI 👻: Bringing Horror Stories to Life with Local Gemma AI"

ch Thriveni — Sat, 31 Oct 2026 09:13:23 +0000

👻 What I Built

What if a machine could turn your imagination into a chilling horror story?

I built GhostWriter AI, a local AI-powered horror story generator that transforms simple story ideas into mysterious, suspenseful and terrifying tales.

Inspired by the Ghost in the Machine theme, I wanted to explore how open-weight AI could bring creative storytelling to life without depending on a paid external AI API.

Users can:

Enter their own horror story idea.
Choose from different horror themes.
Select story length.
Adjust the fear level.
Generate original horror stories.
Download generated stories as TXT files.

🎬 Demo

The application runs locally using Streamlit.

Application Interface

Generated Horror Story

💻 Code

GitHub Repository:

https://github.com/Thrive-123/GhostWriter-AI

⚙️ How I Built It

I built GhostWriter AI using:

Python – Application logic.
Streamlit – Interactive user interface.
Ollama – Running the model locally.
Gemma 3 1B – Open-weight language model for generating stories.

The application sends the user's story idea and selected preferences to Gemma 3 1B through Ollama. The model generates a story, which is displayed in the Streamlit interface.

🌍 Why Does Open Innovation Matter?

Open innovation was important to this project because I wanted to experiment with generative AI while keeping the application simple and accessible.

Using Gemma 3 1B through Ollama gives GhostWriter AI several advantages:

Local inference: After downloading the model, stories can be generated locally without sending prompts to an external AI API.
Privacy: User prompts can remain on the local machine.
Cost control: No per-request external AI API charges.
Flexibility: The model can be changed or experimented with through Ollama.

This project helped me understand how open-weight models can power practical AI applications.

👻 The Spirit of It

GhostWriter AI is my interpretation of the Ghost in the Machine theme.

It explores the idea of a machine bringing imaginary worlds to life through storytelling.

A simple prompt becomes a mysterious tale, and the next story is always waiting in the shadows.

🏆 Prize Categories

Best Use of Gemma

GhostWriter AI uses Google's Gemma 3 1B open-weight model through Ollama for local inference.

Built with curiosity, creativity and a little bit of fear. 👻

Your imagination. Your local AI. Your story.

SellerSpark AI: Building a Local AI Marketing Assistant with Gemma 3 🌟

ch Thriveni — Sat, 17 Oct 2026 15:02:23 +0000

Small-business owners and online sellers often need product descriptions, social media captions, and sales messages to promote their products. Creating this content takes time, especially for people starting a small business or side hustle.

For Hacktoberfest 2026 Week 2: Side Hustle, I built SellerSpark AI, a local AI-powered marketing assistant that helps turn product details into ready-to-use sales content.

🚀 What I Built

SellerSpark AI helps small-business owners, handmade-product sellers, online resellers, and student entrepreneurs create marketing content from a few product details.

Users can enter:

Product name and optional price

Product features

Target customers

Platform: Instagram, WhatsApp, or online marketplace

Language: English or Telugu

Writing style: Friendly, Professional, Creative, or Premium

The application generates a product title, short description, key selling points, a platform-specific sales message, and relevant keywords and hashtags. Users can copy the result or download it as a text file.

📸 Application Demo

Here is a look at the SellerSpark AI interface.

🛠️ How I Built It

I used:

Python for application logic

Streamlit for the user interface

Ollama to run the model locally

Gemma 3 (1B) as the open-weight language model

When a user submits product information, the application builds a prompt and sends it to Gemma 3 through the local Ollama instance. The generated marketing content appears in the Streamlit interface, where the user can copy or download it.

🌱 Why Does Open Innovation Matter?

Using an open-weight model gave me an opportunity to build and experiment with an AI-powered application on my own computer.

With Gemma 3 running through Ollama, the application can generate content locally without requiring a paid hosted-model API for inference. This gives developers more control over experimentation and provides an accessible way to learn how AI applications work.

Local inference still requires compatible hardware and a downloaded model. AI-generated content also needs review because the model may invent product details or unsupported benefits.

💡 Who Could Pay for It?

SellerSpark AI is designed for small-business owners, handmade-product sellers, online resellers, and student entrepreneurs.

A seller who regularly writes product descriptions and social media posts might find value in generating first drafts faster. Whether sellers would pay for it depends on the quality of the results and how useful it is in their daily work.

This is an early prototype, and I have not yet validated willingness to pay with a real customer.

💻 Code and Setup

GitHub Repository: SellerSpark AI

The source code, dependencies, setup instructions, and application demo screenshot are available in the repository.

Run Locally

1. Install the Python dependencies

pip install -r requirements.txt

2. Install Ollama

Download and install Ollama from ollama.com.

3. Download Gemma 3 (1B)

ollama pull gemma3:1b

4. Start SellerSpark AI

python -m streamlit run app.py

Open the local URL shown in your terminal to use the application.

🔮 What's Next?

My next goals are to test the application with small-business sellers, collect genuine feedback, improve the reliability of generated claims, and make the experience more useful for Telugu-speaking users.

Building SellerSpark AI has helped me explore how open-weight AI can be used to address a practical problem. I would love to hear suggestions for making it more useful to sellers.

🏆 Prize Category

I used Gemma 3 (1B) as the model powering this project and intend to enter the relevant Gemma partner category, subject to the challenge's eligibility rules.

🌿 TouchGrass AI: Turning Screen Time Into Outdoor Time with Local Gemma

ch Thriveni — Thu, 08 Oct 2026 08:45:15 +0000

What if AI's job wasn't to keep us on a screen longer, but to help us leave the screen?

That was the idea behind TouchGrass AI, my project for the Hacktoberfest 2026 Week 1 challenge: Touch Grass.

The goal is simple: give AI a small role in planning an outdoor experience, then let the user step away from the screen and actually do it.

🌱 What I Built

TouchGrass AI is a personalized outdoor activity planner.

The user provides:

⏰ Available time
👥 Number of people
😊 Mood
🌳 Interest
📍 Location

The application then uses Gemma 3, an open-weight AI model running locally through Ollama, to generate a personalized outdoor plan.

The plan includes:

A suitable outdoor activity
A time-based schedule
Things to carry
A screen-free challenge
An alternative outdoor activity

The final goal is not another AI conversation.

The goal is to get the user outside. 🌿

🤖 Why Open-Weight AI?

The open part of this project isn't just an extra feature. It is what makes the application work.

I chose Gemma 3 because I wanted the core AI generation to happen locally rather than depending on a closed cloud AI API.

With Ollama, Gemma runs directly on my laptop.

This gives the project a few important advantages:

🔒 Local Processing

The user's preferences can be processed locally on their own computer instead of being sent to a remote AI service for generating the plan.

💰 No Per-Request AI API Cost

Once the model is downloaded, the application can generate plans locally without requiring a paid API request for every interaction.

🔄 Model Flexibility

Because the model is running through Ollama, the architecture can potentially be adapted to use other locally available models in the future.

🌐 Local AI Inference

After the model has been downloaded, the AI inference itself can run without an internet connection.

For a project designed to encourage people to spend less time online, having the AI itself run locally felt especially appropriate.

🏗️ How It Works

The workflow is intentionally simple:
User Preferences
↓
Streamlit Interface
↓
Python Application
↓
Ollama
↓
Gemma 3
↓
Personalized Outdoor Plan
↓
📵 Screen-Free Challenge
↓
🌳 Real-World Activity

🛠️ Tech Stack

Python
Streamlit
Ollama
Gemma 3 (1B)
Git
GitHub ## 📸 What the App Looks Like ### 🏠 User Input

🌳 Generated Outdoor Plan

📵 Screen-Free Challenge

🔗 Project

The complete source code is available on GitHub:

TouchGrass AI — GitHub

Built for the Hacktoberfest 2026 Week 1: Touch Grass challenge.

devchallenge #hf26challenge