{"slug": "noxara-touch-grass-look-up-and-let-the-universe-tell-you-a-story", "title": "Noxara: Touch Grass, Look Up, and Let the Universe Tell You a Story", "summary": "A developer built Noxara, an offline-first mobile stargazing companion that identifies celestial objects using device location and orientation rather than camera image recognition, drawing on the Yale Bright Star Catalogue's 9,110 entries. The app runs Gemma 3 270M locally in the browser via Transformers.js with WebGPU or a WebAssembly/CPU fallback to generate short astronomical narratives, which the open-source Piper neural text-to-speech engine then narrates aloud. The project, built with React, TypeScript and Vite, is intended to let users listen while looking at the sky instead of depending on a remote inference API.", "body_md": "*This is a submission for the [Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass](https://dev.to/challenges/hacktoberfest-week1-2026-10-05)*\n\n**An offline-first AI stargazing companion that wants you to spend less time looking at your screen and more time looking at the sky.**\n\nWe've built an incredible amount of technology designed to keep us looking down.\n\nAnother notification. Another feed. Another reason to spend five more minutes scrolling.\n\nMeanwhile, there's an entire universe above us.\n\nI built **Noxara** around a simple question:\n\n**What if AI could encourage us to spend more time experiencing the real world instead of pulling us further into our screens?**\n\nNoxara is a mobile-first stargazing companion that helps you identify celestial objects by pointing your phone at the night sky, then uses on-device AI to tell you their stories.\n\nIt's a little astronomy guide, a little storyteller, and hopefully, a reason to go outside tonight.\n\nOkay, you can't actually touch the stars.\n\nBut you can leave your room, find somewhere quiet, and look at them.\n\nStargazing is one of those rare activities that gets better the farther you get from civilization. Away from streetlights, buildings, and light pollution.\n\nUnfortunately, those are also places where a reliable internet connection isn't guaranteed.\n\nThat shaped one of Noxara's most important design decisions:\n\n**The AI shouldn't need a server to tell you about the stars.**\n\n**1. Go outside and look up**\n\nPoint your phone toward something interesting in the night sky.\n\nNoxara uses your geographic location, device orientation, and astronomical calculations to identify celestial objects in the direction you're pointing.\n\nNo camera-based image recognition is required.\n\n**2. Discover what you're seeing**\n\nThat tiny point of light might be a star hundreds of light-years away, a planet, or something that has guided travelers for centuries.\n\nNoxara uses the Yale Bright Star Catalogue, containing 9,110 entries, alongside enriched astronomical metadata to provide context about your discovery.\n\n**3. Let the universe tell you a story**\n\nThis is where AI comes in.\n\nRather than displaying only a catalogue of measurements, Noxara uses a locally running Gemma model to generate a short narrative based on astronomical context.\n\nPiper text-to-speech then narrates it aloud.\n\nThe idea is simple: **listen while looking at the actual sky instead of reading paragraphs on your phone.**\n\n**4. Stay curious**\n\nWant to know why a star has a particular name? How far away it is? Why ancient civilizations considered it important?\n\nAsk a follow-up question and let your local AI guide respond.\n\nThe demo walks through the initial model installation, identifying a celestial object, generating a story, and listening to it through Piper.\n\nNoxara is designed for mobile browsers, where location and motion sensors make the experience possible.\n\n**GitHub:** [https://github.com/CosmicAlison/Noxara](https://github.com/CosmicAlison/Noxara)\n\nNoxara is built with React, TypeScript, Vite, astronomical calculation libraries, and browser-based AI inference.\n\nThe repository includes the application source, model worker implementations, astronomical catalogue integration, and tests.\n\nMy goal was to create an experience where AI supports an activity happening in the physical world, without depending on a remote inference API.\n\nNoxara uses **Gemma 3 270M** for generating short astronomical narratives and answering follow-up questions.\n\nGemma runs in the browser through Transformers.js, using WebGPU where supported or a WebAssembly/CPU fallback.\n\nInstead of sending every discovery to an external language model service, Noxara constructs a prompt using the identified celestial object and available astronomical facts.\n\nThe model generates its response directly on the device.\n\nFor narration, I use **Piper**, an open-source neural text-to-speech engine that can run locally.\n\nGetting this right was one of the more interesting technical challenges.\n\nMy first implementation used Kokoro. While the generated audio was promising, inference on my Samsung Galaxy A55 introduced long pauses between sentences.\n\nI experimented with chunking and performance comparisons before switching to Piper.\n\nPiper delivered a much more practical narration experience on my device, without requiring an external speech API.\n\nThat mattered because Noxara isn't supposed to feel like another chatbot.\n\nIt's supposed to feel like a companion you can listen to while exploring the night sky.\n\nThe object identification process doesn't depend on an AI image recognition API.\n\nInstead, Noxara combines:\n\nThis provides the factual foundation for Gemma's storytelling.\n\nOn first launch, Noxara guides users through installing Gemma and Piper.\n\nBoth models are initialized during setup, with model resources cached locally for later use.\n\nThe language model and speech engine run in separate workers, helping keep expensive inference operations away from the main interface.\n\n**The current offline boundary is important:** after setup, AI text generation and speech synthesis can run locally without cloud inference. However, reliably reopening the entire web application with no connection still requires additional application-shell caching.\n\nFor Noxara, open innovation isn't just about which libraries I imported.\n\n**It's what makes the product's entire philosophy possible.**\n\nImagine driving away from the city, finding a quiet place with minimal light pollution, and looking up at a sky full of stars.\n\nNow imagine your astronomy guide refusing to work because it can't reach an AI server.\n\nThat would defeat the point.\n\nBy running Gemma and Piper on the device, Noxara can generate narratives and speech without making cloud inference requests once the models are installed.\n\nA closed, API-only approach would require connectivity whenever the user wanted a new story or answer.\n\nOpen models allow me to move that intelligence onto the phone.\n\nLocation is essential to Noxara because the stars visible above you depend on where you are.\n\nBut knowing where someone is stargazing doesn't mean an AI provider needs to receive their coordinates.\n\nNoxara performs its astronomical calculations locally and doesn't need to send the user's location to a cloud language model.\n\nThere's no account requirement, and the AI inference workflow doesn't require a remote server to process the user's questions.\n\nThis is especially meaningful for an application built around exploration.\n\n**Your curiosity about the world shouldn't require giving up your privacy.**\n\nThis was perhaps the most practical benefit I experienced.\n\nWhen Kokoro proved too slow for comfortable narration on my phone, I wasn't locked into a proprietary speech provider.\n\nI could experiment with different open-source implementations, compare performance, and switch to Piper.\n\nThe result was a better experience for the hardware I actually had.\n\nOpen innovation gave me control over the trade-offs between model size, latency, and output quality.\n\nI can continue experimenting with the language model, voice, prompts, and inference strategy without redesigning the application around a vendor's API.\n\nCloud inference APIs introduce a recurring cost every time users generate content.\n\nNoxara's local approach avoids per-request cloud inference charges for its Gemma storytelling and Piper narration.\n\nThat doesn't make the project literally free to operate: hosting, development, initial downloads, and users' device resources still have costs.\n\nBut it changes the economics of experimentation.\n\nI can build and test an AI experience without paying a provider every time someone asks about a star.\n\nAnd for a small, independent project, that freedom matters.\n\nThere's an interesting contradiction at the heart of Noxara.\n\nI'm using AI to encourage people to spend *less* time interacting with technology.\n\nNot another feed. Not another endless conversation. Not another screen demanding attention.\n\nJust enough intelligence to make the physical world a little more interesting.\n\nThat's what excites me about open innovation: the freedom to build AI experiences that don't have to revolve around keeping users connected to a platform.\n\nSometimes the best thing an AI can do is give you a reason to put your phone down.\n\nOf course, I couldn't build an app about touching grass and then only test it from my desk! So one night, I headed out to the west coast of Mauritius to give Noxara a proper field test. I pointed my phone toward the night sky, and Noxara identified **HR 5191**, a star in the Yale Bright Star Catalogue. That star is **Alkaid**, part of the constellation Ursa Major and located at the tip of the Big Dipper's handle. What I find especially fascinating about Alkaid is its history in traditional Arabic astronomy: the stars of the Big Dipper were once imagined as a funeral procession, with Alkaid representing the leader of the mourners. Noxara successfully generated a story for my discovery, although the audio initially failed, which gave me a real-world issue to work on. It was a small but exciting moment: standing outside, looking at an actual star, and seeing something I'd built connect that point of light to thousands of years of astronomy and human storytelling. That's exactly the kind of experience I wanted Noxara to create.\n\nNoxara is still evolving, and there are a few areas I'd love to improve:\n\nOne of my favorite things about astronomy is that the same sky has inspired people across thousands of years.\n\nThe stars have been calendars, navigational tools, scientific mysteries, and characters in stories passed down through generations.\n\nI wanted Noxara to bring a little of that wonder back.\n\nAnd if it convinces even a few people to step outside tonight, find somewhere dark, and look up for a while, I'll consider that a pretty good outcome.\n\n**Touch grass. Look up. The universe has stories to tell.** 🌌\n\n*Built for the Hacktoberfest Open-Source AI Challenge — Week 1: Touch Grass.*", "url": "https://wpnews.pro/news/noxara-touch-grass-look-up-and-let-the-universe-tell-you-a-story", "canonical_source": "https://dev.to/alison_pariela/noxara-touch-grass-look-up-and-let-the-universe-tell-you-a-story-22mm", "published_at": "2026-10-11 17:21:34+00:00", "updated_at": "2026-10-11 17:30:14.955171+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-tools", "generative-ai"], "entities": ["Noxara", "Gemma 3 270M", "Transformers.js", "Piper", "Yale Bright Star Catalogue", "React", "TypeScript", "Vite"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/noxara-touch-grass-look-up-and-let-the-universe-tell-you-a-story", "markdown": "https://wpnews.pro/news/noxara-touch-grass-look-up-and-let-the-universe-tell-you-a-story.md", "text": "https://wpnews.pro/news/noxara-touch-grass-look-up-and-let-the-universe-tell-you-a-story.txt", "jsonld": "https://wpnews.pro/news/noxara-touch-grass-look-up-and-let-the-universe-tell-you-a-story.jsonld"}}