This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass I built GenQuest, a local AI-powered adventure generator that gives people a reason to step away from their screens and experience the world around them.
Tell GenQuest a little about yourself, how much time you have, your interests, and where you're comfortable spending that time. It generates three personalised quests for you to choose from. You might end up observing the sky, exploring your surroundings, creating something, or trying an everyday activity in a new way.
Once you've chosen a quest, the screen takes a back seat. When you return, you can reflect on what you experienced.
Your world. Your next adventure.
I built GenQuest using Python, Ollama, and Qwen2.5 1.5B Instruct, an open-weight language model that runs locally on my laptop.
The user provides their age range, occupation, environment, available time, interests, energy level, and preferred setting. GenQuest sends this context to the local model, which generates three real-world quests for the user to choose from.
Python handles the application logic, including collecting preferences, validating the model's output, checking basic constraints, and providing fallback quests if the model is unavailable or returns invalid output. I used Rich to create the interactive terminal interface.
The AI workflow runs through Ollama on the user's machine. GenQuest doesn't require a hosted AI API, an account, or a database.
Open innovation made it possible for me to build GenQuest around an AI model that runs directly on my laptop, without depending on a paid, cloud-hosted AI API.
This matters because GenQuest is designed to help people spend less time on screens. Running the model locally means quest generation doesn't require sending requests to a remote AI service, and the application can work offline after the model and dependencies have been installed.
Using an open-weight model also gave me the freedom to experiment with prompts, adapt the generation process to different users, and build fallback behaviour around the model's limitations.