Every file, not a sample
Each document is opened and read in full rather than skimmed because the set is large. You get a count at the end: how many were processed and how many failed.
Batch PDF Analysis Tool
Extract the same fields from many papers and reports, then consolidate them into traceable answers, tables, or reports.
Pulling the same handful of fields out of a hundred PDFs and assembling them into one table is not a hard task. It is a task that has to be repeated a hundred times without drifting, skipping, or mixing entries up, and human attention starts sliding somewhere around document thirty.
Batch PDF analysis turns it into a single instruction. You define what to pull from each file; the agent opens every one, extracts, records where each value came from, and consolidates the result into whatever shape you asked for. When a file genuinely lacks a field, it writes "not found" instead of filling the cell with something plausible.
What it can do
You define the fields
Sample size and effect size from papers, three specific expense lines from filings, key clauses from contracts. Fields can be added later without restarting the job.
Every value is traceable
Each extracted item points back to the file and passage it came from, so spot-checking is quick. That matters most when the numbers feed a decision.
Gaps reported honestly
When a document really does not contain the field, it is marked as not found with a reason rather than filled with a reasonable-looking value. This is the failure mode in bulk extraction that is hardest to catch on your own.
How it works
- Upload and parseFiles are parsed one by one after bulk upload, with text recognition applied to scans. Anything that fails to parse is listed separately rather than silently folded into the results.
- Pin down the criteriaYou state the fields and the rules for judging them. Where the definition is ambiguous the agent asks first, instead of guessing a rule and applying it to all hundred files.
- Extract with provenanceValues are pulled against one consistent standard, and the location of each value in the source is recorded as it goes.
- Consolidate and deliverOutput arrives as a spreadsheet, a comparison report, or cited answers. Tables can be downloaded and used directly.
Worked examples
Turning a paper set into a table
For these 120 papers, extract sample size, intervention, primary outcome, and effect size for each, and give me one Excel sheet.
A 120-row table, one row per paper, with missing values explicitly marked, ready for downstream synthesis.
Comparing filings across an industry
Here are annual reports from 15 companies in the same sector. Compare how their R&D spend as a share of revenue moved over three years.
A comparison report with a table and trend commentary where each figure links back to the relevant passage in a specific filing.
Frequently asked questions
How many files can I upload at once?
Batches in the hundreds. Large sets are processed in stages with per-file results retained, and you get totals for processed and failed documents when it finishes.
Does it handle scanned PDFs?
Yes, scans go through text recognition first. Pages where recognition quality is poor are flagged for you to confirm rather than passed downstream as if they were reliable.
How accurate is the extraction?
Every value carries its source, so spot-check a few. More useful than a headline accuracy figure is that gaps are reported as gaps, which lets you see which cells need human attention instead of being misled by a fully populated table.
Can results be exported to Excel?
Yes. Consolidated tables download directly, and you can specify column order and header format.