Hey folks, hope everyone's data pipelines are flowing smoothly today.
I've been testing Perplexity's new 'Upload' feature for a quick-and-dirty way to pull insights from some vendor data sheets and whitepapers (PDFs). It's super handy for summarizing text, but I've hit a consistent snag: it's absolutely mangling any tabular data.
For example, I uploaded a standard PDF pricing sheet with a clean, simple table. When I asked Perplexity to list the tiers and their limits, it returned a jumbled mess. The rows were merged, columns were swapped, and some numerical data was just plain wrong. It seems like it's trying to interpret the table as plain text and losing all the structure.
Here's a snippet of what it gave me vs. what the PDF actually contained:
```text
// What Perplexity returned (paraphrased):
"Starter plan: 10K MTD events, 3 seats Pro plan: 50K MTD events, 10 seats Enterprise plan: Contact sales, Unlimited seats"
// Actual table structure in PDF:
| Plan | Monthly Tracked Events | Seats |
|------------|------------------------|-------|
| Starter | 10,000 | 3 |
| Pro | 50,000 | 10 |
| Enterprise | Custom | Unlimited |
```
It's like it's doing an OCR or text extraction that ignores the visual layout. This is a dealbreaker for any kind of data analysis work.
I'm curious if others are seeing this? Has anyone found a workaround, like preprocessing the PDFs into CSV first? I love the concept for text, but for anything with structured data, it's currently more misleading than helpful. Maybe they'll improve it with better PDF parsing libraries.
ship it
ship it