The latest articles on DEV Community by Zihan Zhao (@zihan64). https://dev.to/zihan64 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4209841%2F2e490794-6fa6-45c7-aa8e-f318046928d7.png https://dev.to/zihan64 en Zihan Zhao Tue, 27 Oct 2026 03:40:35 +0000 https://dev.to/zihan64/hikingpsu-clear-trail-ratings-and-an-ai-guide-for-happy-valley-27h8 https://dev.to/zihan64/hikingpsu-clear-trail-ratings-and-an-ai-guide-for-happy-valley-27h8 <h2> What we built </h2> <p>State College has a strong hiking tradition, and great trails start just outside town. But information about them is scattered across club pages, forums, PDFs and word of mouth. On popular apps, a trail is labeled only Easy, Moderate or Hard, with no reason given. Details that matter here are often missing, too, like whether you can get there without a car or whether it's hunting season.</p> <p>At HackPSU, our team of four built <strong><a href="https://hikingpsu.com" rel="noopener noreferrer">HikingPSU</a></strong>: one place where Penn State hikers can find trails around Happy Valley, see clear ratings, and build a hiking community together.</p> <p> <iframe src="https://www.youtube.com/embed/6LZJRBs1ZFU" width="710" height="399"> </iframe> </p> <h2> How it works </h2> <ul> <li> <strong>A map of 52 trails.</strong> We imported 40 from OpenStreetMap and entered 12 by hand.</li> <li> <strong>A five-level difficulty scale that explains itself.</strong> Effort = miles + feet of climb ÷ 500. Rough terrain bumps a trail up a level. Every rating shows its reasons, so "Level 3" means something.</li> <li> <strong>An AI trail planner.</strong> Users ask hiking questions, and our AI agent, powered by the Gemini API, answers using our trail data.</li> <li> <strong>Check-ins and badges.</strong> When you finish a hike near a trailhead, you can check in with GPS and earn badges.</li> </ul> <h2> What we used </h2> <ul> <li> <strong>Frontend:</strong> React, Vite, Leaflet and Tailwind, hosted on AWS Amplify</li> <li> <strong>Backend:</strong> FastAPI on Render</li> <li> <strong>Data:</strong> MongoDB and OpenStreetMap</li> <li> <strong>Login:</strong> Auth0</li> <li> <strong>AI:</strong> Gemini API</li> </ul> <h2> What broke at 3am </h2> <ul> <li> <strong>The guest limit didn't work.</strong> Guests get 5 AI questions a day, but in testing they could keep going. We keyed the limit by IP address and read the IP from the last entry in <code>X-Forwarded-For</code>. Behind our host's proxy, that entry was the proxy's own address, and it changed between requests, so every request looked like a new visitor. We switched to Cloudflare's <code>CF-Connecting-IP</code> header, which holds the real visitor's IP.</li> <li> <strong>Trails looked too easy.</strong> Many trails we imported from OpenStreetMap are out-and-back, but we rated them on the one-way distance. Once we rated them on the full round trip, the number of Level 1 trails dropped from 28 to 16.</li> <li> <strong>Merge conflicts hit main.</strong> Four people pushing to one branch at 2am led to conflict markers in committed code. We fixed it and added branch protection with pull-request reviews.</li> <li> <strong>The chatbot printed raw asterisks.</strong> The AI replied in Markdown, but the chat window showed <code>**</code> and <code>-</code> as plain text. We added light Markdown rendering.</li> </ul> <h2> What's next </h2> <p>We want to grow HikingPSU from Happy Valley to the rest of Pennsylvania. The same OpenStreetMap import works for any region, and every check-in helps verify the trail data. If you're in State College, try it at <a href="https://hikingpsu.com" rel="noopener noreferrer">hikingpsu.com</a>.</p> mlhacks devchallenge hackathon