Whiteknuckle
I've driven a 46ft fifth wheel and 22ft long truck almost every week for the last 6 months. As of this writing that's about 12k mostly unremarkable, fairly easy miles... mostly. When things are tight in a rig that size, the drives can be very stressful. There are apps out there for RV navigation, and they're mostly pretty good, but they don't tell you the difference between a road you can fit on and a road you shouldn't fit on.
Finally, after 6 months of bad directions and mostly good luck, I decided that I could build an app that would help me review and "drive" my route before deciding if I could take it. I shared the initial Claude concept in a FB group, and learned I was for sure not alone.
This app is the solution. It doesn't just look at length and height, it also looks at wheelbase, pin location, bed length, turning radius, engine power, and then generates a 3d model and report showing where you'll drive the road, where the tight turns are, where the bridges will be tight, and where the shoulders will be non-existent.
It's become a much bigger idea now, and I've got quite a few people signing up to test the early versions. It's a difficult app to design, but I've already found it really useful for a few routes, and hope that I can make it accurate and helpful for others.
Driving New Priest Grade
- Oct 20 – Oct 22
- 30 prompts
- 246M tokens
- +8,303 −115
This is the route that kicked off the whole project. Highway 120 from the central CA valley up to Yosemite. My RV nav app said "go for it". My eyes said "woah that's tight". It is. Now it's turned into this whole project.







Planning a route
- Oct 21 – Nov 3
- 113 prompts
- 627M tokens
- +8,129 −1,857
These are the first steps I took -- just seeing if I could use the app to generate a route. It was all based on the first prototype and showed me a lot of things that needed work, like how the app "drove", how it determined what a good or dangerous turn was, how roads rendered, etc.




























Tunnels, Bridges, and Turns
- Oct 21 – Nov 2
- 169 prompts
- 1.2B tokens
- +15,093 −1,717
The easy part was putting the rig model on the map. The hard part was making the map drivable. To do that, I've had to make custom tooling for tunnel entrances and profiles, bridge trusses and length, roundabout dynamics, lane congruency, interstate interchanges, road lines, you name it.





























































The drive
- Oct 21 – Nov 2
- 83 prompts
- 846M tokens
- +8,394 −1,260
There's a lot more here than putting in some depth data and getting the scene to load. I have to bring in data from each state, from different satellite providers, then I have to simulate it with the real models, and load in the textures while keeping it performant. Lots of weird quirks and edge cases.




















A generated world
- Oct 22 – Nov 2
- 73 prompts
- 1.1B tokens
- +8,692 −993
I started just assuming I'd be able to overlay satellite data, but it's not so easy. Performance is an issue, reliability is a problem, and the main thing is that a flat image looks terrible on contours when there's bridges, interchanges and rivers. I've now switched to building out my own generated version of the world with trees, climates, and crops.



































The rig
- Oct 22 – Nov 2
- 72 prompts
- 435M tokens
- +8,093 −968
I started with the truck and camper from my portfolio, but now I've got models for all Super Duty trucks in every trim, and every variation of Brinkley camper with a few different hitch types.















The UI
- Oct 22 – Nov 2
- 95 prompts
- 659M tokens
- +8,724 −1,786
The app UI has been more involved than I thought. This is more than a routing app, but there's a ton of detail that I want to give to the user. It's also a map + simulator + AI chat + turn by turn... there's a lot of simplification that I still have to do.
























The homepage (and brand)
- Oct 22 – Nov 2
- 138 prompts
- 596M tokens
- +2,353 −170
I asked Claude to generate about 50 variations of a brandmark. First time I've done this before. It took a while but the one I chose jumped right out.
I also wanted to make the homepage a three.js scene, but without all of the scene bloat, so I precompiled these with a limited camera viewing angle and made them work like an Instagram reel on mobile.



















