Grain.AI

Grain.AI measures a rice or paddy sample from a photograph taken on the tray in front of you. It counts the grains, works out what share of them are broken, and reports the length-to-width ratio of the sample — then records the result with the lot, the centre and your name against it.

It is a screening and record-keeping tool. It measures what a camera can see, and it is explicit about the rest.

WHAT IT MEASURES

• Grain count — the app separates touching grains itself, so you pour the sample into a single layer and photograph it. You do not have to spread the kernels apart by hand. Where grains lie on top of one another and cannot be told apart, the app says how far the count may be understated rather than presenting a figure it cannot stand behind.

• Broken share — using the Bangladesh standard's own definition: a kernel counts as broken when it is shorter than 3/4 of the average whole-kernel length in that same sample. Because the rule compares grains against each other, it needs no ruler in the frame.

• Confidence interval — every broken percentage comes with a 95% range, not a bare number. A few hundred grains cannot pin a percentage down tightly, and the app says so instead of implying a precision it does not have.

• Length-to-width ratio — the shape measure that separates slender grain from bold.

• Chalk indicator — flags grains with an unusually bright, opaque area. Front lighting only reveals obvious chalk, so this is a flag to follow up, not a chalk percentage.

• Discoloured indicator — flags grains that are markedly darker or markedly yellower than the other grains in the same photograph. Heat damage and mould show as dark or amber. It is judged against the sample itself under the same light, not against a standard, so it is a flag to follow up, not a damaged-grain percentage.

WHAT IT DOES NOT MEASURE

• Moisture. A phone camera cannot see water content — the light that reveals it is infrared, outside what any phone sensor records. Enter the reading from your moisture meter and the app stores it beside the photograph, together with which meter you used.

• Grain age, and fortified kernels. Both need laboratory instruments, not a camera.

• Boro, Aman and Aus. These are growing seasons, not anything visible in a kernel. You select the season from the docket; the app records it, it does not guess it.

HOW IT KEEPS THE RECORD HONEST

• Every capture is saved, including the ones you retake. If a lot was photographed six times before an acceptable number appeared, the file shows that.

• Bad frames are refused rather than measured. If the image is soft, glared, or the grains are piled on top of each other, the shutter stays disabled and tells you what to fix.

• A photograph you already have can be measured too, and it faces the same checks as a live capture. A soft or crowded one is turned down either way.

• Nothing is compared against a tolerance until you say whether the tray is parboiled or raw rice. The two are graded to different limits, so the app asks rather than assume.

• Export the whole log as a CSV file and share it wherever you need it.

• The app measures. The grading decision, and the signature on it, stay with you.

PRIVACY

The app never uploads a photograph by itself. Photographs stay in the app's own private storage on your phone, and one leaves only if you export it yourself through your phone's share menu, to somewhere you choose.

There are no accounts, no advertising and no third-party analytics.

One feature can use the internet, and it is switched off when you install the app. If you switch on Improve accuracy, the app can send six settings it fitted to your own hand counts, plus a few averages, so results from different centres can be pooled. No photograph, no name, no lot or centre number, no identifier belonging to your phone. You can read the exact text before you send it, and switch it off again at any time.

English and Bangla.