Weekly Status Update: August 15th, 2026
Hit a wall with the maps data pipeline — going back a step to build a stronger foundation, retraining detections on 320 courses, and starting to productize Fleet Orchard.
Sometimes going forward in a single direction and finding incomplete answers is hard to take. This week I hit a wall with my data pipeline that was only going to cause more pain and suffering going forward. I have a path forward, but it feels defeating to go backwards a bit to build a stronger foundation.
Let’s get to updates.
Maps
I’ve been building out the pipeline so that any additional info I can glean — like scraped scorecards — informs and corrects my detections. What I found, though, is that it’s difficult to consolidate or correct maps when I have over-detections on many key classes (looking at you, greens).
My original dataset was 50 hand-annotated golf courses, and I trained on that. Using active learning and more annotations, I’m now up to 320 golf courses. I’ve kicked off training — currently at 35/100 epochs. Here’s to hoping. Promising results thus far.
Side Projects
Spending a good amount of time waiting for App Review. 2 of 3 apps are still in review.
Started productizing Fleet Orchard — the technology I built to power Dewy Marshal and Clay. Going well thus far, hoping to release it for free/paid folks in a week or so.