Whiteboard OCR that actually understands the board.

OCR isn't the hard part

Basic OCR — character recognition from an image — has been solved for decades. The hard part is what you do with the text once you have it.

A whiteboard isn't a document. Text on a whiteboard has spatial meaning. A word inside a circle means something different from the same word written in a list. An arrow between two items asserts a relationship. A crossed-out line records a rejected idea.

Plain OCR flattens all of that. You get words in approximate reading order — and no sense of what the board actually said.

How BoardSnap's OCR layer works

Apple VisionKit handles the character and word recognition layer. It runs on-device, uses the iPhone's Neural Engine, and handles handwriting well — including mixed case, varied pen widths, and the slightly-hurried script of a real meeting.

VisionKit returns recognized text with position and confidence data. BoardSnap uses the positional data to understand spatial relationships — not just "what words are on the board" but "where are they and what does their position mean."

That spatial awareness feeds directly into the AI summarization step.

From characters to structure: what BoardSnap AI adds

OCR gives you characters. BoardSnap AI gives you comprehension.

The AI model receives the recognized text with spatial metadata and interprets it as a document with structure: headings, lists, tasks, decisions, relationships. It identifies which items are action items, which are context, and which are decisions already made.

The output isn't raw OCR text with newlines. It's a structured summary — titled, sectioned, and followed by a task list with states and subtasks. Something you can actually use.

• Handwriting recognition via Apple's Neural Engine — runs on-device
• Spatial relationship parsing — arrows, boxes, clusters all carry meaning
• Task extraction distinguishes action items from context and decisions
• Subtask generation expands compound tasks into discrete steps

What kinds of whiteboards does it handle

Printed text and typed notes: Easy. VisionKit reads clean printed handwriting reliably.

Cursive and fast handwriting: Works well on modern iPhones. Accuracy improves when letters are connected but legible. Very rushed handwriting may have a word or two misread — the summary context usually self-corrects.

Diagrams with text labels: BoardSnap AI reads the labels and infers their structural role from position.

Mixed content — sticky notes, printed text, handwriting together: All processed together. The spatial metadata helps distinguish regions.

Dry-erase boards under fluorescent lights: Common scenario. VisionKit handles the glare better than a manual photograph because it uses edge and contrast data rather than relying purely on color.