Solving the Traveling Salesman Problem* - More efficient route planning with LLMs

* April Fools

In my future career as a school bus driver (part-time hours, benefits, summers off!), I want to efficiently pick up and drop off students. The tool below should help accomplish that goal. You can use this CSV file  to try it out.

I used Claude's Artifacts to put this little tool together, then used DeepSeek R1 to clean it up. Claude 3.5 Sonnet is supremely capable (I made this before 3.7 came out), but I was thinking DeepSeek R1 could outperform it. After asking it to optimize in various ways, Claude's algorithm still had the fastest estimated route times for various datasets. Not super scientific I know, but I'm planning bus routes, not performing rocket surgery. 

It's not exactly solving the traveling salesman problem, but it's sorta good! I'd win a Nobel prize if it did. Or would Claude win the award? Either way, using only addresses, it creates a decent route based on direction of travel. Obviously, staying above 50mph would be the fastest method but that's harder for the students to jump off. 

Since it just spits out a link to Google Maps, you can drag/drop the order as needed. If little Jimmy isn't at school that day, you can just remove his stop before you start navigation. As a nice bonus, you can add more than ten stops which Google doesn't allow by default, which is why morethan10.com exists. I keep finding undocumented Google features the more I vibe code these things.

As I wrote this, I realized the local school buses pick up/drop off students at bus stops and not individual houses. That being the case, I'd be curious to see how this could optimize bus stop routes. Regardless, I enjoy making these ¯\_(ツ)_/¯