Project overview · verified data engineering
From price-list images
to auditable data.
A document-processing pipeline for Karachi fruit and vegetable price lists, with provenance, stable identities, review records, and a read-only public data application.
The business problem
Price-list images are difficult to query, compare, and publish consistently. The system turns document inputs into structured, reviewable outputs while preserving source identity and separating private processing data from public exports.
The repository documentation explicitly excludes private images, credentials, QA, histories, model files, and frozen benchmark artifacts from the public release.
System flow
Prepare
Use document-specific geometry, OCR, category profiles, and table contracts.
Resolve
Assign stable identities from normalized category and source bytes.
Review
Write final and review records with explicit QA and rollback behavior.
Publish
Export selected public data to a read-only React/Vite PWA.
Verified scope
Auditable processing,
with clear limits.
The repository contains Python pipeline code, production-data contracts, static export tooling, unit and regression tests, and public dashboard screenshots. Google Vision, PaddleOCR, and optional local Qwen fallback are documented dependencies; document accuracy still depends on approved QA and source data.
Source: public repository README, architecture, status, deployment docs, and screenshots.
A useful pattern
Make messy inputs
reviewable.
Document processing becomes more useful when identity, provenance, QA, and public output are designed together.
Discuss a data workflow ↗