openfoodfacts/openfoodfacts-ai
This is a tracking repo for all our AI projects. 🍕 🤖🍼
What's novel
This is a tracking repo for all our AI projects. 🍕 🤖🍼
Code Analysis
13 files read · 4 roundsA collection of experimental AI/ML pipelines for Open Food Facts covering ingredient extraction (NER), logo classification, LLM-based spellchecking, LLM evaluation frameworks, and nutrition table detection, each as independent sub-projects with their own training/inference code.
Strengths
The ingredient extraction pipeline shows genuine depth with a custom token classification pipeline fork, domain-specific prompt engineering for spellcheck, and robust LLM-based data generation with hallucination validation. The LLM evaluation framework is well-structured with proper caching, assertion-based scoring, and multi-provider support.
Weaknesses
The repo is a hodgepodge of 15+ independent sub-projects with no shared abstractions, inconsistent dependency management (some use poetry, some uv, some requirements.txt), and minimal test coverage. Several projects (nutrisight, logo-classifier) are thin wrappers around HuggingFace examples or simple linear classifiers with limited originality.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
60
Contributors
19
Files
450
Active weeks
22
Repository
Language
Python
Stars
265
Forks
70
License
—