Skip to content

Topic

Document Data Extraction

Pulling structured fields such as prices, areas and dates out of web pages and PDFs with language models in the loop.

Current stories

buildOne report1 publisher

Catching a model-swap regression takes twenty labeled documents run twenty times each

A dev.to write-up swaps the single assertion for per-field hit rates with 95 percent intervals over repeated trials. On its own example table, the pair it calls a real regression has intervals that overlap by 0.3 points.

Publishers:dev.to

Reality

Evidence30
Adoption
Not scored
Hype gap+35
Incentives75
Confidence60
buildOne report1 publisher

Identical 0.95 confidence on right and wrong answers killed Scribe's threshold gate

Nine local VLM configurations read the same 23 handwritten clinical pages against 221 hand-curated gold values, and accuracy on the fields accepted without human review rose from 75% to 96% once a second model started checking the first.

Publishers:dev.to

Reality

Evidence58
Adoption8
Hype gap−10
Incentives45
Confidence58
buildOne report1 publisher

AWS's Quick Automate walkthrough turns a plain-language prompt into an RFI workbook-to-CSV workflow, promoted via Import/Export

AWS's tutorial builds an RFI questionnaire automation in seven steps, from an S3 action connector to an import into production. When the workbook format changes, the documented fix is to edit the instructions.

Publishers:aws.amazon.com

Reality

Evidence52
Adoption
Not scored
Hype gap+30
Incentives88
Confidence62
buildOne report1 publisher

Velrim's null-field benchmark catches its own API inventing 11% of absent fields

The pooled 17% belongs to a table with a 40% outlier in it, so the part that transfers is the scoring rule: a mechanical null check against fields someone has verified are missing, plus a blank-rate column to keep it honest.

Publishers:dev.to

Reality

Evidence62
Adoption12
Hype gap−8
Incentives80
Confidence55
buildOne report1 publisher

Eight percent of fields left empty beats two percent quietly wrong

A dev.to argument for lifting numbers out of documents by pattern before the model ever sees them rests on a cost asymmetry between gaps and wrong digits, and on a schema guarantee that covers shape rather than truth.

Publishers:dev.to

Reality

Evidence38
Adoption
Not scored
Hype gap+10
Incentives45
Confidence47