novarki AI Solutions Studio

Case study · Document, vision & knowledge intelligence

Is this safe for me to eat?

One question, answered in seconds, from a photo of a label. AllergySnap turns ingredient lists into a clear safety signal for a person's own allergens — and explains why.

The problem

For a family managing food allergies, every packaged product is a reading exercise under time pressure: dense ingredient lists, synonyms for the same allergen, “may contain” advisories, labels that change between batches. The FDA’s nine major allergens are only the start; many people react to ingredients no standard list covers. A grocery aisle is a poor place to make a safety decision from fine print, and existing apps typically require an account, an internet connection and a fixed list of allergens.

What we built

AllergySnap reads the product instead of the person. Scan a barcode or QR code, or point the camera at an ingredient label, and the app automatically works through its scan methods without asking the user to pick a mode. The answer is one of four explicit signals — Safe, Warning, Danger, Unknown — with the matching ingredients highlighted, so the person can see the reason, not just the verdict. Profiles cover the FDA Big 9 plus any number of custom allergens. Results can be shared with caregivers and family by text, QR code or link.

AllergySnap is free. It works offline, using cached products and US Government nutrition data, and it requires no account: the person’s allergen profile stays on their phone.

How it works

Barcode, QR or label photo in; Safe · Warning · Danger · Unknown out — one element in accent: the signal.

Barcode and QR scans are always free and unlimited. Photo scans of labels are handled by vision AI and proprietary matching logic that maps ingredient names and their synonyms to the person’s allergen profile. When the app cannot be confident, it says Unknown and tells the person to check manually, rather than guessing.

Design decisions

An explicit “Unknown”. The most important safety decision in the product. A system that always produces a confident answer is dangerous in this domain; the honest fourth state is what makes the other three trustworthy.

Explain every flag. The highlighted ingredients are the product. A verdict without a reason cannot be checked by the person holding the box, and this is their body.

Offline and account-free by design. Grocery stores have bad signal; allergy data is personal. Both constraints were treated as requirements, not features.

Automatic scanning. People in an aisle do not want to choose between barcode mode and photo mode. The app progresses through methods itself.

Outcome

AllergySnap is live on the App Store and is used by families managing food allergies. The full feature set — Big 9 plus custom allergens, offline operation, caregiver sharing — is available without an account.

What we learned

In safety-adjacent products, the quality of the “I don’t know” matters more than the quality of the “yes”. We spent more effort on when to withhold an answer, and on how to show the evidence, than on raw recognition accuracy, and it is the reason people trust the app with their children’s food.

What this means for your project

Any workflow where a person has to make a decision from a document, a label or a form — intake paperwork, compliance checklists, invoices, inspection photos — benefits from the same pattern: read it, match it against the rules that matter to this person or this business, show the evidence, and be explicit about uncertainty.

Tell us what you're trying to do.

We reply within two business days with a clear view of whether and how we can help.

Start a conversation