A homework-photo app found the hard part is not OCR but date ambiguity and task splitting
Hayk_D · reddit · 2026-07-21
A handwritten-homework photo app turned out to be mostly a parsing problem
The author built an app that lets students photograph homework, whiteboards, or worksheets and turns them into structured tasks with fields like subject, due date, and estimated effort.
The pipeline is straightforward but tricky in practice:
- image goes in
- Claude’s vision API reads it
- a prompt extracts structured fields into JSON
- the result is shown as an editable task card before saving
The post says vision accuracy was not the main issue. The hard parts were:
- ambiguity: “due Friday” depends on the current date, so the app had to pass the date into the prompt and reason about the nearest Friday
- confidence handling: low-confidence parses are flagged for review instead of silently saved
- multiple assignments in one image: the app had to detect and split them rather than merging them into one broken task
The author asks for others’ experiences with photo parsing and vision models.
More from Multimodal
- Solo founder turns complaints on screen into bug reports with a local MCP server — phdptsd · 2026-07-22
- Google demo says Gemma 4 can inspect car damage from video in under 6 seconds — soumitrashukla9 · 2026-07-22
- SenseNova-U1-8B adds local text and layout editing without losing image quality — SandyL925 · 2026-07-22
- Sonilo Sound Effects 1.0 goes live on fal with video- and text-to-audio generation — JaynitMakwana · 2026-07-22
- MIT's New Algorithm Speeds Up AI Diffusion Sampling, Wins ICML Outstanding Paper — MIT_CSAIL · 2026-07-22
- Local image-to-video releases around LTX 2.3 appear to be slowing down — EasternAverage8 · 2026-07-22