This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass
What I Built
Modern life traps our attention behind glowing rectangles, doomscrolling through feeds while losing connection with the living world around us. WildSense is an offline-first progressive web application built to reverse screen fatigue by turning every outdoor walk, community trail, and backyard into an interactive naturalist expedition.
Key Capabilities
- Ask Naturalist AI ("Found Something New?"): A dedicated field assistant designed specifically for botanical, ornithological, mycological, and entomological questions. WildSense enforces strict ecological guardrails—if an inquiry is non-nature related (like asking about smartphones or coding), it gently steers the explorer back to the living environment.
- Photo Classification Lab: Upload real-time trail photos or inspect specimen samples. The engine extracts visual features to return taxonomic classification, distinguishing field traits, habitat context, and strict educational safety protocols.
- Field Trait Matcher: Built for explorers when camera conditions are poor or lighting is low. A diagnostic keying system lets users select kingdom domains (Flora, Birds, Insects, Fungi, Tracks), tap diagnostic trait chips (lobed leaves, acorn cups, orange breast, gill pores), and cross-examine specimens against professional field checklists.
- Offline Field Journal: Every encounter, species classification, GPS tag, and naturalist note is stored locally in the browser's IndexedDB via Dexie.js. Your observations belong to you and persist without requiring external cloud accounts or constant server connections.
- Micro-Adventures & Quests: Gamified field missions—from canopy audits and urban bird counts to fungi foraging trails—designed to get families, students, and solo adventurers off the couch and onto real trails.
Demo
- Live Deployed App: https://wildsense-3eje.onrender.com
- Mobile First Design: Optimized for single-handed mobile navigation on phones, with full offline caching support for remote trail excursions.
Code
The entire project is open-source under the MIT license:
🌿 WildSense — An Offline-First AI Nature Companion
"Less screen time. More world discovered." Built for the "Touch Grass" Open-Source AI Hackathon.
WildSense is an AI-powered outdoor exploration companion designed to help people disconnect from notifications and reconnect with the natural world through guided, mindful outdoor missions. Unlike traditional apps that keep your eyes glued to the glass, WildSense prepares an outdoor experience, tracks your immersion, and encourages you to slip your phone into your pocket.
✨ Key Features
🧭 1. Adaptive Outdoor Expeditions & Missions
- Tailored Exploration: Choose from 6 rich domains (Birds, Plants & Trees, Insects, Forests & Woodlands, Mindfulness, and General Exploration).
- Flexible Duration & Pacing: Pick from 5-minute sensory breaths, 15-minute neighborhood strolls, 30-minute deep walks, or 60-minute wilderness expeditions.
- Difficulty Modes: Beginner (gentle observation), Intermediate (active tracking), and Adventurous (deep immersion).
- Sensory Observation Tips…
How I Built It
WildSense is engineered around a resilient dual-layer architecture tailored for outdoor environments where connectivity fluctuates:
- Local-First Data Layer: Built using Dexie.js (IndexedDB) for zero-latency, offline persistence of journal entries, checklists, and specimen records.
-
Multimodal AI Integration: Leverages Google Gemini 1.5 Flash (
@google/genai) with structured JSON schema outputs to deliver structured taxonomic cards, confidence ratings, and ecological notes. - Frontend Stack: Built with React 18, TypeScript, Tailwind CSS, and Vite, accompanied by Lucide React iconography and mobile-responsive viewport layouts.
- Strict Nature Boundary Guardrails: Prompt engineering protocols filter out off-topic text and non-biological queries to keep the companion centered strictly on outdoor discovery.
Why Does Open Innovation Matter?
When hiking deep in a national park or backcountry trail, cellular reception drops to zero. Proprietary, cloud-only AI assistants fail completely the moment you lose signal, locking knowledge behind paywalls and remote servers.
Open innovation matters because:
- Nature Exploration Must Be Universal: Field naturalists, students, and hikers shouldn't need expensive subscription tiers to learn the species living in their local biome.
- Privacy by Default: Field locations and personal journal entries remain local on the user's device rather than being harvested for tracking or advertising.
- Resilience in the Field: Open-weight models and local-first software guarantee that your tools keep working miles beyond the reach of cell towers.
My Agent Session
This project was developed, iteratively debugged, tested in the browser, and deployed with the assistance of Antigravity and DevRelay:
How can we build an offline-first nature exploration companion that gets people off screens and into the wild, with on-device/local AI identification and strict error detection for non-nature objects?
We'll architect WildSense around an offline-first design:
1. Micro-missions with minimal screen time that generate outdoor observation cards and printable Pocket Guides.
2. An on-device botanical and zoological identification lab combining multimodal LLM vision (Gemini/Gemma/Ollama) and local taxonomy matching.
3. Strict semantic validation that distinguishes real outdoor wildlife from non-nature/artificial objects with clear error reporting.
4. Dexie.js IndexedDB storage ensuring 100% offline functionality in backcountry trails without cellular coverage.
Implement the identification lab so that real nature queries (like bird plumage, tree bark, mushrooms) give scientific breakdowns, but unknown text or artificial items (like cars, phones) return clear errors.
Implemented the dual LLM + offline identification pipeline in IdentificationLab.tsx, fallback-provider.ts, and gemini-provider.ts. Validated with live tests across birds, fungi, insects, plants, and non-nature detection.
Prize Categories
- Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass (Main Track)
What's Next for WildSense
- On-Device Edge Models (WebGPU): Running quantized lightweight vision models (like MobileNet or Small Vision Transformers) directly in-browser for 100% offline image inference without API dependencies.
- Audio Bioacoustics: Adding bird song and insect stridulation identification via microphone recordings on trail walks.
- Community Trail Maps: Exporting local observations in standard GeoJSON format to contribute citizen-science data directly to platforms like iNaturalist and GBIF.
Conclusion & Gratitude
WildSense was created with one simple conviction: technology should bring us closer to our physical world, not isolate us from it.
Thank you to DEV, the open-source community, and the Hacktoberfest organizers for encouraging AI projects that motivate people to step away from their desks and touch grass! 🌲🌱
Top comments (1)
Some comments may only be visible to logged-in visitors. Sign in to view all comments.