Tahini is sesame: teaching a small open model to spot hidden allergens
Description and use
BoldPrint automatically generates allergen‑highlighted labels, nutrition tables, and Nutri‑Score, reducing missed allergens to 3 % and costing $0.12 per thousand labels.
Problem
Small food sellers need accurate allergen labels but manual labeling is error‑prone and costly.
Outcome
BoldPrint reduces missed allergens from 74% (untrained model) to 3% (shipped product) on 32 real recipes, achieving $0.12 per thousand labels.
Workflow
- Paste recipe text
- Model parses ingredients and identifies allergens
- Cross‑check with Open Food Facts lexicon
- Generate nutrition table and Nutri‑Score
- Export PNG or PDF sticker
- Optional human review of ⚠ Check lines
Evidence
On 32 real-world recipes I wrote and labelled by hand (brands, typos, several languages): ... BoldPrint as shipped (model + lexicon cross-check) 3.0% missed allergens, 2/32 labels with a miss.
The lexicon alone missed 11% of the allergens, and 9 of the 32 labels had at least one miss.
Over 20 RL steps, missed allergens per label on the training batches fell from about 0.05 to under 0.02, and the average reward roughly tripled (0.22 → 0.66).
- Record date: 2026-10-17T10:41:20.647Z