Project note, August 2026: This post records a capture decision made while building Seasoned Pan.
Food logging fails at intake.
The calculations can be perfect and the charts can be polished, but none of that matters if recording a familiar jar, package, or snack requires transcribing a nutrition label field by field.
The obvious fashionable answer was image recognition. Point a camera at the food, ask an AI model what it sees, and let the software estimate the nutrition.
I chose barcodes first.
Identification Is Not Measurement
An image model may recognize a plate as pasta. It may even make a plausible guess about the sauce. It cannot reliably see how much oil went into the pan, how large the serving is, which brand was used, or whether an ingredient was substituted.
That makes photo-based nutrition a chain of estimates presented through a very confident interface.
A packaged-food barcode has a narrower job. It identifies a product. Seasoned Pan can use that identifier to request structured product data, normalize the response, and present the result for human review.
The user still verifies the label. The software simply removes most of the typing.
A Free Database Was Good Enough To Start
The lookup uses Open Food Facts, a community-maintained food database. Its records are not perfectly uniform, so the application cannot blindly copy whatever comes back.
The normalization layer prefers per-serving values when they are available, falls back to per-100-gram values when appropriate, and maps inconsistent upstream fields into the same form used for manually entered foods.
The lookup runs through the site rather than directly from the browser. That gives the request a consistent identity and keeps messy third-party data handling in one place.
Camera Support Still Matters
Barcode-first does not mean keyboard-only.
On browsers that expose barcode detection, the phone can open its camera and read the code directly. On devices without that capability, the number can still be entered manually. The fallback is part of the feature, not an apology attached to it.
That distinction matters because device support is uneven. A capture workflow should degrade into a little more typing, not a dead end.
AI Can Wait For A Better Job
There may eventually be a useful place for image recognition in the project. It could suggest a food name, identify likely ingredients, or create a rough starting point for an unlabeled meal.
If added, it should be visibly labeled as an estimate and should never overwrite verified nutrition data.
The barcode decision was not anti-AI. It was pro-specificity. For packaged food, a product identifier and a structured database do more useful work than a broad visual guess.
The best automation is not always the most impressive demonstration. Sometimes it is the method that quietly removes thirty seconds of irritating data entry and leaves the user in control.