Microclimate Runner

Microclimate Runner: A Local-AI Route Planner That Gets You Outside

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

Microclimate Runner is an open-source running route planner designed to help people stop staring at route maps and actually go outside.

The app uses your current location, target distance, and preferred priority to find nearby routes. It looks for routes with more shade, lower pollen exposure, and better overall comfort.

The flow is intentionally short:

1. Choose a distance and route preference.
2. Let the app plan a nearby route.
3. Read or listen to the briefing.
4. Start Run Mode and leave the screen behind.

While running, the app tracks progress with GPS and displays only the essential information needed to keep moving.

Demo

• Live demo: https://microclimate-runner.vercel.app/

Demo Notes

The local development setup uses Gemma through Ollama for route ranking. The hosted Vercel demo uses the application's explicit deterministic fallback when local Ollama inference is unavailable, because a Vercel deployment cannot access Ollama running on my personal computer.

The interface reports the scoring source instead of claiming that the hosted version used Gemma when it did not.

Code

• GitHub repository: https://github.com/divyyansh-063/microclimate-runner

How I Built It

The project is built with Next.js, React, TypeScript, OpenStreetMap data, Open-Meteo weather and air-quality data, and a Mastra workflow.

The open-source AI component is local Gemma running through Ollama. Gemma receives route candidates, runner preferences, and environmental conditions, then ranks the candidate paths for comfort.

The main workflow is:

1. OpenStreetMap/Overpass provides nearby footpaths and route candidates.
2. Open-Meteo provides current temperature, UV, wind, and pollen data when available.
3. Local Gemma ranks the routes using an inspectable prompt.
4. A deterministic heuristic scorer is used as an explicit fallback if Ollama or Gemma is unavailable.
5. Optional ElevenLabs narration creates a short audio briefing.
6. Run Mode uses browser GPS so the runner can put the phone away.

The application never pretends that local AI was used when it was unavailable. The interface reports whether the result came from:

• gemma-local
• heuristic-fallback

To run the project locally:

bash
npm install
npm run dev

## Prize Categories

I am entering the overall Hacktoberfest Week 1: Touch Grass challenge.

The project uses local open-source AI through Gemma and Ollama. I am not claiming a partner category unless the challenge page lists a category that this project genuinely qualifies for.