README.md

metadata

license: mit
language:
  - en
pretty_name: Open Food Facts Canada Final Deliverables
tags:
  - data-quality
  - deduplication
  - open-food-facts

Open Food Facts (Canada) — Final Deliverables

Welcome to the offCanada Final Deliverables repository. This repository contains the complete dataset processing pipelines, deduplication models, data quality migration frameworks, and regulatory rule-packs developed for the Open Food Facts Canada initiative.

---

📁 Repository Overview

This repository is structured into two main independent deliverable folders:

offCanada/Final_Deliverables/
├── OpenDB_Compliments_Merged/   # Deduplication & Product Grouping Pipeline
├── OFF_DataQuality/              # Perl-to-Python Data Quality & Multi-Engine Framework
└── README.md                     # Top-level navigation guide (this file)

---

🚀 Quick Navigation Guide

1. OpenDB_Compliments_Merged/

Purpose: Deduplication & Canonical Product Mapping for Open Food Facts and Compliments brand product datasets.

• What it does: Groups duplicate product records, generates canonical product representations, and maps raw product listings to canonical clusters using LLM-assisted and candidate generation techniques.
• Key Files to Look At:
  • README.md — Detailed pipeline documentation and workflow instructions.
  • Deduplication_Pipeline_Documentation.docx — Full methodology report and architecture details.
  • canonical_products.parquet — Resulting dataset of canonical/deduplicated products.
  • product_mapping.parquet — Mapping table linking raw product IDs to canonical cluster IDs.
  • deduplicated_comparison.csv — Comparative analysis of original vs. deduplicated product entries.
  • candidate_gen.py & llm_grouper.py — Core Python scripts for candidate cluster generation and LLM product grouping.

---

2. OFF_DataQuality/

Purpose: Legacy Perl-to-Python Data Quality Migration, Multi-Engine Benchmarking (dbt / Soda), and Canadian Food Regulatory Rule-Packs.

• What it does: Modernizes legacy Open Food Facts Perl data quality checks (.pl) into Python validation routines, dbt-core SQL tests, and SodaCL YAML contracts. Features statistical confidence scoring, automated semantic guardrails, and an interactive Streamlit comparison dashboard.
• Key Files to Look At:
  • README.md — Comprehensive technical documentation, architecture diagram, and execution guide.
  • dashboard/app.py — Streamlit interactive dashboard for side-by-side engine comparison and rule metrics.
  • validation/engine_comparison.py & validation/parity_validator.py — Multi-engine parity benchmarking and statistical confidence evaluation engines.
  • rulepacks/registry.py — Rule profile registry (Global rules, Canada-specific food regulatory rules with legal citations, and Hybrid mode).
  • perl_checks/rules/ — 19 legacy Perl data quality rule scripts (.pl).
  • results/engine_comparison.json — Pre-computed multi-engine benchmark comparison report.

---

📌 Summary Table

Folder	Core Focus	Key Output / Entry Point
OpenDB_Compliments_Merged	Product Deduplication & Canonical Mapping	canonical_products.parquet & Deduplication_Pipeline_Documentation.docx
OFF_DataQuality	Data Quality Migration & Benchmarking	dashboard/app.py & validation/engine_comparison.py

For detailed setup, execution commands, and implementation details for each component, please navigate into the respective folder and refer to its internal README.md.