Comment by akoboldfrying
Comment by akoboldfrying 5 days ago
Thanks! I know what you mean about "countless months" on an algorithm... Glad you were able to improve it to a useful level.
> it was more like getting stuck for a while then getting small break throughs
Like simulated annealing itself ;-) (Yes, I watched the deep dive clip)
Really cool that you give API examples right on the landing page BTW. Do you know if many customers use the API directly? Also really interesting to me that your algo is quick enough that you can return a solution directly before the HTTPS connection times out (I would have thought it might take half an hour to optimise, meaning you'd probably need to instead return some kind of handle that the user can use to pick up the solution via a different endpoint later).
A couple of minor nits I noticed, hope this is helpful rather than annoying:
- The "Blog" link at the top appears to produce a copy of the original landing page. Maybe just delete that link?
- The "API Response Structure" section is duplicated
- In the deep dive clip, at 5:31, the subtitle contains "three dpaq.ing", while at 5:54, both the voice and the subtitle say "third bin packer" (I suspect you meant "3D bin packer")
Yes, most of the project evolved because of potential customers who later disappeared. Someone shows up and says, “Hey, we’ll pay you if you provide this feature,” so I implement it… and then they vanish
The API came about the same way. Currently, there’s just one active API user: a company from Sri Lanka that built a Unity WASM interface on top of it.
I coded the algorithm completely from scratch, with no external inspiration and fun fact, it’s written in F# using functional programming. There are a lot of immutable lists floating around. Honestly, I probably wouldn’t have been able to write it without functional programming (even though it’s notoriously slow for this kind of problem). The immutability allowed me to focus on individual parts of the app without worrying about side effects.
That said, how it works and how fast it works still feels like magic to me, just like any meta-heuristic algorithm applied to combinatorial problems.
And thanks for the feedback! I knew the Blog link was broken after I redid the landing page, but I didn’t notice the duplicated API Response Structure section. I’ll definitely fix that. Though part of me thinks, “Should I really bother?”
Also, thanks for catching the audio/subtitle issues. I was aware of both, but I figured it wouldn’t make much difference at this stage. Maybe that attitude explains my low traction… who knows? Either way, I appreciate you taking the time to point these out!