Oriel — AI & technology worth noticing, every day.

A small, not-so-smart model went from 10% to 99.6% at reading property deeds for $1.15, and it says 'check this line' instead of bluffing

Post · https://dev.to/advikdivekar/a-small-not-so-smart-model-went-from-10-to-996-at-reading-property-deeds-for-115-and-it-says-490k

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady N 86°29'...

First seen by Oriel on .

AI summary

DeedReader combines a Qwen3-8B LoRA reader with deterministic validation to parse synthetic property deeds, flag inconsistencies, and plot parcels, reaching 99.6% exact accuracy for about $1.15.

  • The model outputs compact structured calls, while plain code handles bearings, curves, closure, area, unit conversion, plotting, and exports to DXF, GeoJSON, or SVG.
  • Five checks must agree: parsing, 1:5,000 closure, sentence-level number and landmark grounding, and an independent regex parser.
  • On 3,000 fresh synthetic deeds, 2,988 were exact; 10 of 12 imperfect outputs were caught, while two reflected equivalent generator labels.
  • The system supports offline execution through a 369 MB adapter, but curves are chord-form only, real-deed accuracy is unmeasured, and licensed surveyors remain necessary.

#qwen3-8b #lora #property deeds #offline ai

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Reactions

Total public reactions reported by the source; distinct from points and likes.

  • DEV Community: 83, latest observation ; +58 over observed period since

Comments

Comment count reported by the source.

  • DEV Community: 37, latest observation ; +29 over observed period since

Aliases

SourceExternal IDURL
DEV Community4916473https://dev.to/advikdivekar/a-small-not-so-smart-model-went-from-10-to-996-at-reading-property-deeds-for-115-and-it-says-490k Read discussion