The latest articles on DEV Community by Darzi Anas A Jaweed (@anasdarzi009). https://dev.to/anasdarzi009 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4170122%2Fc3135095-4bf8-416c-bc17-96a2b3dca144.png https://dev.to/anasdarzi009 en Darzi Anas A Jaweed Fri, 09 Oct 2026 17:09:18 +0000 https://dev.to/anasdarzi009/trailmate-ai-taking-local-gemma-ai-into-the-real-world-c69 https://dev.to/anasdarzi009/trailmate-ai-taking-local-gemma-ai-into-the-real-world-c69 <p><em>This is a submission for the <a href="https://dev.to/challenges/hacktoberfest-week1-2026-10-05">Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass</a></em></p> <h2> What I Built </h2> <p><strong>TrailMate AI</strong> is an outdoor micro-adventure generator powered by locally running Gemma 3 4B.</p> <p>It transforms simple preferences, such as available time, activity type, difficulty, and environment, into outdoor missions with checkpoints and a grounding exercise.</p> <p>The goal is simple: use AI to give people a reason to put their phones away and engage with the world around them.</p> <p>I tested TrailMate AI in a real park and completed all three generated checkpoints: Observe the Canopy, Listen to the Wind, and Feel the Ground. The app recorded 3/3 checkpoints completed and awarded 125 points.</p> <h2> Demo </h2> <p><strong>Project demonstration:</strong> </p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fpoe9acp9vrcxk40pfdi2.jpeg" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fpoe9acp9vrcxk40pfdi2.jpeg" alt=" " width="800" height="1078"></a></p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffh99i7y67gbqnmzkvr9w.jpeg" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffh99i7y67gbqnmzkvr9w.jpeg" alt=" " width="800" height="533"></a></p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fic0vrt9pjile7r1r9t80.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fic0vrt9pjile7r1r9t80.png" alt=" " width="800" height="500"></a></p> <p>The demo shows TrailMate AI generating an outdoor mission using locally running Gemma and presenting its checkpoints.</p> <h2> Code </h2> <p><strong>GitHub repository:</strong> <a href="https://github.com/Anasdarzi009/trailmate-ai" rel="noopener noreferrer">https://github.com/Anasdarzi009/trailmate-ai</a></p> <p>TrailMate AI is open source under the MIT License.</p> <h2> How I Built It </h2> <p>TrailMate AI uses <strong>Gemma 3 4B</strong>, an open-weight language model, through <strong>Ollama</strong> for local inference.</p> <p>The application is built with:</p> <ul> <li> <strong>AI model:</strong> Gemma 3 4B</li> <li> <strong>Local inference:</strong> Ollama</li> <li> <strong>Backend:</strong> Python, FastAPI, and Pydantic</li> <li> <strong>Frontend:</strong> React, Vite, and Tailwind CSS</li> <li> <strong>Mission history and points:</strong> Browser localStorage</li> </ul> <p>The React frontend sends mission requests to the FastAPI backend, which communicates with the local Ollama runtime. The generated mission is then returned to the interface.</p> <p>Once the model is installed, mission generation does not require a hosted AI API. Initial setup and model downloads require internet access.</p> <h2> Why Does Open Innovation Matter? </h2> <p>Open-weight AI made it possible to build TrailMate AI around local inference instead of depending on a hosted AI API for every mission.</p> <p>This gives developers greater control over where inference happens and provides a foundation for privacy-conscious, locally powered applications.</p> <p>It also makes the project easier for other developers to explore, modify, and extend using open-source tools.</p> <p>For this project, open innovation supports the central idea: AI can help people engage with the real world without making constant online interaction a requirement.</p> <h2> My Agent Session </h2> <p>This section is optional. Add a DevRelay session link if you have one to share; otherwise, remove this section.</p> <h2> Prize Categories </h2> <p><strong>Best Use of Gemma</strong> — Gemma 3 4B is the core model used to generate outdoor missions, running locally through Ollama.</p> <h2> What's Next? </h2> <p>Potential improvements include richer mission personalisation, additional outdoor activities, and more ways to record and track adventures.</p> <p>These are future ideas, not features claimed as already implemented.</p> <p>Built for Hacktoberfest 2026 Week 1: <strong>Touch Grass</strong> 🌿</p> <h1> devchallenge #hf26challenge #ai #opensource*This is a submission for the <a href="https://dev.to/challenges/hacktoberfest-week1-2026-10-05">Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass</a>* </h1> <h2> What I Built </h2> <h2> Demo </h2> <h2> Code </h2> <h2> How I Built It </h2> <h2> Why Does Open Innovation Matter? </h2> <h2> My Agent Session </h2> <h2> Prize Categories </h2> devchallenge hf26challenge ai opensource