How does the filter understand page context?

The filter analyzes the layout and semantic meaning of the text. It can distinguish between a question, a solution, and a table of contents by looking at the page structure and language patterns.

Can it filter by diagrams?

Yes, you can prompt the filter to keep pages that contain diagrams or pages where a question references a diagram on another page.

The Brain Behind the Filter

Why semantic analysis beats keyword searching every time.

When I first started working with large PDF sets, I realized that searching for a keyword like "Trigonometry" wasn't enough. You'd get the questions, sure, but you'd also get 50 pages of marking schemes and cover sheets that happened to mention the word.

The filter in Parse & Pack is different. It doesn't just "find words"—it reads intent.

How it Reads Your Mind

The filtering process happens in three layers:

1. Structural Mapping

We don't just see text; we see blocks. We identify where headings are, where lists start, and where a diagram might be referenced.

2. Semantic Analysis

The tool understands that "Calculate the value of x" is a question, while "The value of x is 5" is a solution. This allows it to exclude answer keys with near-perfect accuracy.

3. Context Retention

One of the most frustratring things about PDF tools is when they cut off a question halfway through. Our filter understands when a question spans multiple pages and keeps the whole unit together.

Pro Tip: Negative Constraints

The best way to refine your filter is by telling it what not to do. If you find yourself getting too many pages, try adding a negative constraint to your prompt:

"Keep pages with geometry diagrams, but strictly exclude the formula sheets and the index pages."

Semantic Filtering FAQ

Does it work with hand-written text?

Matching is based on the text extraction. If your PDF has a high-quality OCR layer for handwriting, it will work well. If not, the accuracy may drop.

Can I filter by page color or images?

The tool primarily focuses on text and layout. While it can detect when an image is likely a diagram, it doesn't "see" the image content in the same way it reads text.