What was the biggest challenge?
Discussion
Performance. I could get the entire globe rendered down to to z18, but it still took a day or two.
What stack did you use? Is the product still around?
It was for a startup called TallyGo https://www.crunchbase.com/organization/clearpath. We had highly accurate route-finding. More accurate than Google when we started (they got better in the subsequent years), but much less capacity to serve requests because we were doing time sensitive A* and not smart approximations. That was Scala on the backend, Spark pipelines and the A* was in Rust.
The entire product was an SDK for mobile developers to integrate into their apps. After 3 years of development we got worried that our map tile provider was going to cut our lunch and build the same thing, so I started working on this tile generation pipeline. The data was from OSM and spark was involved, as was some binary that I don’t recall. Lots of running, waiting, tweaking, re-running etc.
The routing engine work in the prior 3 years was somewhat more interesting and I spent a lot of time on turn-by-turn instructions, which seems simple until you start to get into it. The start-up ultimately failed due to shenanigans when we tried for acquisition. (We got screwed). All the code is shelved away in some holding company now. No regrets though.