inWell demoLegacy well-file extraction & search All demos

About this demonstration

What it is. inWell is a prototype by InLabs LLC that reads scanned legacy well records (well completion forms, driller's logs, well histories — typed and handwritten), extracts a structured record with per-field confidence and provenance (page and verbatim snippet), runs QC rules, and cross-checks the result against the public KGS master list. This site shows a processed sample of 22 documents. It is a public-data demonstration, not an operational tool; every output requires expert review.

Data sources

Method

Pages are rasterized and read by OCR/layout analysis (Azure AI Document Intelligence) and a language model (GPT-4.1, temperature 0, strict JSON schema); every model call is logged. Values are normalized (dates, feet, casing fractions, Sec-Twp-Rge), checked by rules (API format 15-CCC-NNNNN, date order, depth consistency, location ranges, elevation reference order, value present in its own snippet, duplicates) and compared with the KGS master list. Accuracy figures are measured against that list, not claimed.

Measured results

Over all 60 processed documents (126 pages), of which this site shows a sample: 59 were matched to a well in the KGS master list, and 44 carry at least one flag for expert review. Field-by-field agreement with the master list, where both values are present:

FieldMatchMismatchAgreement
api_number580100%
completion_date451476%
elevation_ft41295%
lease_name58198%
location48198%
operator56395%
spud_date43688%
total_depth_ft55493%
well_number58198%

Depth within ±10 ft, elevation within ±3 ft on the same reference (KB, DF or ground), names fuzzy, dates exact. A mismatch can be a reading error or a master-list error; either way it needs expert review. Completion dates often differ because a form and the master list can record different events.

What it does not do

Provenance

All code and design are InLabs' own work. The only data used are public records of the Kansas Geological Survey. No employer, client, partner or vendor documents, code, designs or SDKs and no third-party proprietary material were used. © 2026 InLabs LLC.

Version 0.1.0 · published 2026-11-08.