I've been building an app that lets students snap a photo of an assignment, a whiteboard, a printed worksheet, whatever, and turns it into a structured task with subject, due date, and estimated effort. Want to admit that I went in underestimating how hard this would be from an actual parsing pipeline POV. The pipeline is roughly: photo in → Claude's vision API reads the image → a prompt asks it to extract specific structured fields (title, subject, due date, estimated effort) → returned as JS...