@etohq/workflows-input-dataset-pdf-runtime

Node-only PDF text extraction + user-defined structuring into tabular records.

This package is intentionally separate from @etohq/workflows-input-schema-runtime because PDF parsing is not cross-platform.

Typical Usage

1. Extract PDF text:

import { pdfToText } from "@etohq/workflows-input-dataset-pdf-runtime"
const res = await pdfToText(pdfBuffer)
if (!res.ok) throw new Error(res.error)

2. Structure into records using a user-provided PdfExtractSpec:

import { structureRecordsFromText } from "@etohq/workflows-input-dataset-pdf-runtime"

const structured = structureRecordsFromText(res.text, extractSpec)
// structured.records: Record<string, string>[]

3. Feed those records into dataset import mapping:

import { importCsvRecords } from "@etohq/workflows-input-dataset-runtime"

const imported = importCsvRecords({
  schema,
  spec: datasetImportSpec,
  records: structured.records,
})

User-Defined Structure (Regex)

The core idea is: users define a regex with named capture groups that represent columns:

Example line regex:

^(?<S\\/N>\\d+)\\s+(?<Appnum>\\w+)\\s+(?<Surname>[A-Z-]+)\\s+(?<Firstname>[A-Z-]+)\\s+(?<Othername>[A-Z-]+)\\s+(?<Program>.+?)\\s+(?<List>Merit|Supplementary)$

Then map the group names to output column names via column_map.