Field notes
Building Raychis in the open.
A 10-part series on building a real-world, on-device plant identifier from scratch - what worked, what broke, and what turned out to be much harder than it looked. A new part lands at 8am every Wednesday, with the last arriving the day the app goes live.
4 of 10 published
-
Plant Identification Looks Solved. Until You Try to Build It
An app told me a toxic plant was basil. I spent a year building a plant identifier from scratch to understand why - and where computer vision quietly breaks.
-
The First Thing That Broke Was the Data
Before the model, the data. Corrupt images, mislabelled species, and the unglamorous pipeline work that decides whether anything downstream can work at all.
-
The Data Looked Fixed. The Model Knew Better
The class distribution finally looked right - and the model still failed. What clean-looking data hides, and how training surfaces the problems you thought you had solved.
-
The Commit Before the Silence
The last commit of 2025, then fifty-two days of nothing. What stalls a solo build, and what it takes to come back to a project you have walked away from.
-
The Two Weeks I Stopped Building
-
The Data Valley
-
The Reframe
-
The Breakage
-
The Rule I Set for Myself: The Stage Gate
-
The App That Will Tell You It Might Be Wrong
The series is published first on Medium; these are the same pieces, kept here for the record.