This is a submission for the Hacktoberfest Open-Source AI Challenge: Week 1 — Touch Grass.*
Tags: #devchallenge #hf26challenge
What I built
Trailhead — a single-file web app that turns a trail name, distance, season, and group into a hiking brief: gear checklist, safety brief, Leave No Trace reminders, and a go/no-go verdict. The screen time is under two minutes. Everything after that happens outside.
The whole thing is one HTML file. No build step, no server, no framework. Open it on your phone in the trailhead parking lot and it works.
The demo
You fill in five fields — trail, distance, season, who's going, anything else — and hit "Plan my hike." Thirty seconds later you get a practical brief. Example: Bear Mountain Loop, 3–7 miles, fall, 2 adults + 1 kid → layered clothing call, headlamp reminder (shorter days), kid-specific snack and turnaround-time math, and the one risk that matters most for that trip.
The prompt engineering is the product: the model is instructed to be specific, not generic — no "bring water" fluff without quantities, no safety boilerplate without the actual risks of this hike.
Why open-weight is the point
Trailhead calls Llama 3.3 70B — Meta's open-weight model — through Groq's free tier. That choice is load-bearing, not decorative:
- It costs nothing to run. A hiker checking trail conditions shouldn't need a subscription. Free-tier inference on an open model means the app is free forever, for everyone.
- No vendor lock-in. The model is one string in the code. Swap in Mixtral, Qwen, or DeepSeek by changing a single line. If Groq changes terms, point it at any OpenAI-compatible endpoint.
- Auditable advice. When a model tells someone whether a hike is safe, you want to know what model said it and be able to swap it. Open weights mean the safety advice isn't a black box owned by one company.
- It runs from anywhere. One static file. No backend to deploy, no API keys baked in (yours stays in your browser). A closed model behind a paywalled API couldn't do this at $0.
Closed AI would make this a $9.99/month app with an account wall. Open weights make it a file you can text to a friend.
What I'd do next
Offline mode with a small local model (Llama 3.2 1B via WebGPU) for trails with no signal — the parking lot has bars, the summit doesn't. And a "trip report" loop: hikers confirm or correct the brief afterward, building a dataset of real-world accuracy.
Try it
The app is a single trailhead.html — open it, paste a free Groq key from console.groq.com, plan a hike, then touch grass. 🌲
Top comments (0)