Document inspection

Fraud Detection

A SvelteKit workspace for running OCR on document images, reviewing confidence-based signals, and checking annotated regions in context.

0 Fixtures

0 OCR words

0 Regions

Loading fixtures...

Repository state

Project overview

Fixture documents

Loaded from src/lib/assets/documents

OCR words

Parsed from cached OCR JSON and uploads

Annotated regions

0 marked forged in metadata

Input formats

JPG / PNG

Accepted by the upload endpoint

Signals

What the app checks

OCR confidence

Words below confidence thresholds are grouped with deterministic review labels.

Annotated regions

Fixture metadata can mark document regions as suspicious or forged for overlay review.

Manual verdict

The final forged/genuine state is a reviewer action, not an automatic legal decision.

Pipeline

How it runs

Load an image

Start with a JPG or PNG upload, or open one of the tracked document fixtures.

Run OCR

Tesseract.js extracts words and confidence values through SvelteKit API routes.

Review signals

The UI groups low-confidence OCR words and annotated regions for manual inspection.

Fixtures

Tracked documents

Loading fixtures...

Implementation

Project notes

SvelteKit workspace

UI routes, API endpoints, and stores live under src/.

Tesseract.js OCR

Uploads are processed through the local /api/ocr endpoint.

Document fixtures

Sample images and OCR JSON files are tracked with the app.

Deterministic scoring

Severity comes from OCR confidence and anomaly metadata.

Run locally

Analyzer workspace

Use the analyzer to validate fixtures or uploaded images.

The canvas, OCR confidence groups, annotated regions, and verdict controls stay together for repeatable local testing.