📄 August 2026 Platform Update: The PDF Experience, Rebuilt

Hi there,

Nexadata has turned PDFs into data for over a year. What it has not had is an experience worthy of the capability. Extraction worked, but everything around it asked too much of you, and the path from a document to a usable Dataset had more friction in it than it should have.

This release rebuilds that path end-to-end. PDF is now a Data Format handled like any other, the table review workflow has been redesigned around speed, and the rough edges are gone. The capability is not new. The experience is.

📄 A Rebuilt PDF Experience

PDF is now a Data Format. It sits on the Connect Data step alongside Tabular and Spreadsheet, in the same radio group, so you create a PDF Dataset exactly the way you create any other one. Selecting PDF adds a Process PDF step to the wizard, which opens a dedicated workspace covering extraction, table review, and building.

Table Review, Redesigned Around Speed

Turning a document into a Dataset means making a decision about every table inside it, and a dense document detects more tables than you expect. Footers, address blocks, and signature panels all read as tables to an extraction engine. The new Table Selection stage is built so that getting through them is fast.

• One click per table. Keep table or Not relevant. The same gesture on table one and on table two hundred.
• Recognize a table without opening it. Each entry in the rail shows its page and shape, such as p.1 and 3 rows × 2 cols.
• Never lose your place. Kept tables get a green check and discarded ones a red cross, with Pending, Approved, and Rejected filters and a reviewed counter. Filter to Pending and work down to an empty list.
• Set column types in the same pass. The Column Editor sits on the right of the same screen with a type dropdown per column, so you type a table while its contents are in front of you.
• Rename and rotate in place. Type over the name field, with no edit mode and no separate save. Flip header blocks whose labels run down the side upright with Rotate.

Every control does one thing, sits where you expect it, and gives you feedback immediately. That is the whole point of this release.

Setup Happens Once Per Document Format

Every decision you make is recorded as a template. The next invoice, statement, or report in the same layout runs against it without repeating the review, because what the template depends on is the layout rather than the values. New numbers in the same tables are exactly the designed case.

That is why the review speed matters. A 600-page private equity quarterly report is a realistic input, and Nexadata stitches every table that spans a page break across the whole document. Curating it is real work the first time and close to free every quarter after.

The example in our documentation is a legal invoice, chosen because it is awkward in all the ways real documents are, but nothing here is specific to legal work. Supplier and freight invoices, bank and brokerage statements, utility bills, remittance advices, shipping manifests, lab results, insurance schedules, fund and board reports, regulatory filings. The source file showed you totals. The Dataset gives you the arithmetic behind them, per line, ready to aggregate any way you need.

📄 Setting Up a PDF Dataset →

🔢 Column Data Types in the Dataset Builder

A column of hours or amounts read as text is effectively inert. You cannot sum it, average it, or group by it, which blocks the first thing most people want to do with extracted data.

The Dataset Builder now infers column data types automatically and lets you override the inference whenever it gets one wrong. A Hours column that came through as text becomes numeric, and the Group By with a Sum aggregation you were reaching for simply works.

🚫 Reverse Filters

The filter transformation could only express inclusion. Every operator was affirmative, so there was no clean way to say something as ordinary as "drop the rows where Account starts with 9." You had to build around the gap rather than through it.

Filters now express exclusion directly, with negated operators sitting alongside the existing ones. On extracted documents in particular, where subtotal and summary rows land in the middle of your detail, this is the difference between a clean Dataset and a fiddly one.

🧱 Platform Fixes

• ENG-994 · Bug Fix: Tables marked as rotated were not automatically joined to the primary table, so their values stayed separate instead of appearing alongside the records they qualify. Rotated tables typically hold context such as matter name, invoice number, period, or entity, and those values are meant to repeat across every record. Automatic joining now works as intended.

🔭 What's Next

We are turning the "set it up once" idea into a first-class platform capability and broadening where your files can live:

• Reusable Workflow Templates: Save a whole workflow as a configuration file, with variables and automatic column backfill, then stand the same solution up again without rebuilding it
• Workflow-Level Variables: Variables scoped to a workflow rather than the workspace, with values set at template import so a template stays portable across tenants
• Native OneDrive and Google Drive Connectors: Read and write files directly, with live connections bound to the SSO identity you already sign in with
• Format-Aware File Browsing: Choosing a Data Format will scope the file list to matching files, so picking a PDF no longer means scrolling past everything else

Thanks for being part of the Nexadata journey!

The Nexadata Team