Dark Sky Walk

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

What I Built

Dark Sky Walk: an offline night-sky companion that gets people out of the house after sunset.

Most stargazing apps keep you staring at the phone, and they break down where it matters most: dark spots with no signal. Dark Sky Walk flips that. Before you leave, you ask in plain language, "What can I see tonight, and where should I go?" The app works out what is visible (planets, the Moon phase, bright stars, any meteor shower peaking this week), checks cloud and moonrise timing, and suggests a short walk to a darker nearby spot with a simple "look here, then here" checklist you can follow with your phone in your pocket.

Once you are outside, the screen is the shortest part of the experience: a red-light, low-brightness card for each target, plus optional spoken prompts so you can keep your eyes on the sky.

Who it is for: beginners who have never found a constellation, families looking for an evening activity, and hikers or campers who want to make use of clear nights without any signal.

How I Built It

• Local language model: a small open-weight model (for example a quantized Llama, Qwen or Gemma variant) running through a local inference runtime such as llama.cpp or Ollama. It turns questions like "what's good tonight?" into a friendly, ordered observing plan.
• Open astronomy data: an open-source ephemeris library (such as Skyfield or Astropy) with a downloaded star catalog calculates real positions of planets, the Moon and stars for your exact time and location. The model never guesses sky positions; it only explains them.
• Tool-calling agent loop: an open-source agent framework lets the model call the ephemeris, moon-phase and sunset tools, then assemble the answer.
• Offline maps: open map data (OpenStreetMap extracts) to find parks, hills and open fields nearby, plus a light-pollution layer to rank them by darkness.
• Interface: a lightweight web or mobile UI in night mode, with large text, red-tinted colors and optional text-to-speech from an open voice model.
• Field-first design: all data and models are downloaded once at home, so the whole pipeline works on a laptop or phone with no connection.

Why Does Open Innovation Matter?

• It works where the signal doesn't. Dark skies are found far from cell towers. A cloud API fails exactly where the app is needed; a local model keeps working in a field or on a hill.
• Your location stays yours. The app only needs to know where you are standing at night. Running locally means that information never leaves the device.
• It costs nothing to run. No API bills or rate limits for a hobby project that people might use every clear night.
• You can adapt it. Swap in a smaller model for an older phone, fine-tune it to explain things at a kid's level, or change the agent's behavior (for example, favoring meteor showers in August) without waiting for a provider to add a feature.
• The science stays trustworthy. Open ephemeris data does the astronomy, and the model handles only the explanation, so the app stays accurate and inspectable instead of hallucinating where Jupiter is.