Do we need to buy new handhelds?

No. It runs on the Android RF handhelds the warehouse already has. If one is very old we check it first, but the normal case is install and go.

Does the operator have to learn another app?

They sign in with their badge or code, scan the label and shoot. The screen tells them which photo is next. Real training takes fifteen minutes.

What if we do not want AI validation?

Leave it off. It is an option per photo type: you can have the AI on the inbound container and not on the rest.

What if the AI gets it wrong?

It does not decide on its own: the photo is stored with the verdict and the reason, and the warehouse sees it in the Tracker. It is built to be cautious — when in doubt it flags the photo rather than passing it — and if it ever does not answer, the photo is stored anyway.

Who can see the photos?

Whoever signs in to the Tracker with permission to see them. At a logistics operator each customer sees only their own goods; the warehouse sees them all.

Do item photos get mixed between documents?

Only if you want them to. Turn the catalogue on and the photo is shared as the item photo; leave it off and it stays on that document — which is what you want for returns and damaged goods.

How many photos can be taken per item?

As many as you set per type, or no limit: the operator closes when done. If the type updates the catalogue, photographing the same item again replaces the previous set of catalogue photos instead of piling up.

What about receipts already closed?

The photos and their index can be rebuilt from the files on the server: each file name carries the reference and the moment it was taken.

Does this work for outbound, not just inbound?

The module works per document type. Inbound and receipts are the usual ones, but loading a truck is configured the same way.

AI-validated photos · Caotic module in production

Proof of what came off the truck, validated by AI.

The operator photographs the goods with the same handheld they already work with, and an AI validates every photo on the spot: if what must be visible is not, it tells them before the truck leaves. Every photo is linked to the document, the line, the item and the pallet, with the time and the name of whoever took it.

Book a demo How it works

Handheld · AI validationillustration

What it is for

The claim arrives three weeks after unloading. Without photos it is one word against another, and the warehouse ends up paying the difference.

Disputed receipts

A pallet arrives damaged, cartons are missing or the seal is not the one on the delivery note. The dock photo, with date, time and operator, ends the conversation in a minute.

Damaged goods

Whatever arrives damaged is photographed there and then and stays attached to the document line. If the criterion asks for it — “the packaging is intact” — the AI flags the photo as an incident without anyone having to review it.

Product records kept current

An item photo can also go up to the item master and stay there as the catalogue photo, shared by every document. The warehouse ends up with its own photographed catalogue.

How it works on the dock

No second app and no separate phone: it is the usual RF handheld, the one the operator already carries.

Sign in

The operator identifies themselves with their QR badge or by typing their four-digit radio code. The session already knows which warehouse and who they are.

Scan the document

They scan the label of the inbound order or receipt. Caotic checks the document exists before allowing any photo: there are never orphan photos.

Shoot

The screen tells them what each photo must show and displays an example image. Photo 1 of 3, 2 of 3, 3 of 3.

The AI validates

The AI looks at each photo on the spot. If it does not pass it, it asks for a retake; if it does, the photo is stored on the warehouse server and indexed.

Three ways to capture, depending on what you need to prove

Fixed shots

A closed script per capture type: truck with the seal on, doors open, paperwork and seal. In order, with none skipped, and each with its own criterion for the AI.

By item

The handheld lists the items on the document. The operator scans the product — or the pallet —, takes as many photos as needed, and the AI can check that what was photographed is the item in the description.

By line

One photo per document line, with that line description on screen. Useful when the same item arrives on several lines and batches.

And the pallet, if you ask for it

In capture by item, the same scan field accepts the pallet or the SSCC: scanning it also resolves the product. It can be optional or mandatory. From then on each photo is anchored to that pallet and shows up from stock, not only from the document.

AI validation

Every photo type carries a single sentence saying what has to be visible. The AI looks at each photo and checks that sentence before it is stored.

What the AI checks · written by the warehouse

The rear of the container is visible with the doors CLOSED, the security seal fitted and visible, and the number plate of the truck or trailer.

You tell the AI what to look for: Nothing is programmed: the criterion is a sentence in the admin screen, in your own language. “The seal is fitted”, “the packaging is intact”, “it is the item in the description”. Change it today and tomorrow it applies on every handheld.
It warns while there is still time: If the AI sees the photo does not meet the criterion, the operator finds out on the spot, with the truck still in front of them. A retake costs ten seconds; going back the next day is not an option.
It never blocks the dock: If the AI does not answer, the photo is stored anyway. The AI helps the operator; it does not stop unloading.
It leaves a trail: Every photo records whether the AI validated it, whether it passed, the reason the AI gives when it did not, and the model that looked at it. AI incidents are listed together in the Tracker.

Where it all ends up

Photos are consulted from the Tracker, Caotic’s web portal: the warehouse sees them all and each of the operator’s customers sees only their own. Screenshots with demo data.

Every receipt, with its photos

The list of receipts with supplier, reference, container, lines and units. The photo column says how many each one has, red if the AI flagged any. One click opens the receipt with its photos.

Tracker inbound panel with the photo column for each receipt, red when the AI flags an incident — **Inbound**demo data

The receipt, line by line, with its photos

At the top, the photos describing the delivery: truck, seal, paperwork. Below, each document line with its item photos right underneath, not piled up at the end. Green for AI-validated; red for incidents.

Detail of an inbound document: header photos of the container and, under each document line, the photos of that item with the AI validation mark — **Inbound · detail**demo data

Every photo of one document

Type the reference and out come all of its photos: type, line or item, date, operator and AI validation. Click one and it opens full size, exactly as it was taken.

Gallery of the photos of one reference with type, item or line, date, operator and the AI model that validated them — **Photos · by reference**demo data

The photos of a pallet, straight from stock

In pallet-level stock, each row says whether that pallet has photos — red if the AI flagged any. They open from there even if they were taken against another document: the goods carry them.

Pallet-level stock with the photo column and the viewer showing the photos anchored to one pallet, each with its AI validation — **Occupancy · pallet photos**demo data

What the AI did not pass

The history between dates, by type, with thumbnail and operator. One filter leaves only the AI incidents: a single screen shows what is being photographed badly, and by whom.

Photo list with thumbnail, date, type, reference, detail, operator and the result of the AI validation — **Photos · list**demo data

How much has been photographed

Quantities between two dates: captured, AI-validated and incidents by type, and the validations by AI model. This is the figure a logistics operator bills the service on.

Photo summary between two dates by capture type and by AI model, with captured, AI-validated and incidents — **Photos · summary**demo data

Configured, not programmed

Capture types, what each photo must show and what the AI has to validate are edited from the browser, per warehouse. The handhelds pick the change up on their own.

Binari admin screen: photo types for a warehouse, with key, label, document type, mode, example image and the AI validation criterion for each photo — **Admin · photo types**demo data

Capture types: As many as the warehouse needs, with their label and the document type they are scanned against.
What each photo must show: The text the operator reads and an example image showing the framing. In order and mandatory.
Mode: Fixed shots, by item or by line; how many photos per item (or no limit, with the operator closing) and whether scanning the pallet is optional or mandatory.
AI validation: On or off, the sentence the AI checks and the model, photo by photo. In capture by item the criterion can include the item’s real description.
Catalogue: Whether the item photo also goes up to the item master. Switched off for returns and damaged goods, where the photo must stay on that document.

Technical sheet

Devices: The Android RF handhelds the warehouse already uses. No phones or cameras to buy.
Identification: Operator QR badge or four-digit radio code. Every photo records who took it.
Documents: Inbound orders, receipts and whichever document types are configured; the handheld validates the label against Caotic before capturing.
Where they are stored: On the warehouse server, in each company’s photo folder. The file is the original; what the database indexes is where it is and what it is.
What is indexed: Reference, type, slot or line, item, pallet, file, date, operator, AI validation result, reason and model.
AI validation: AI vision models called from the warehouse server: the key never leaves it and the image is sent only for that validation. Switched on, and the model chosen, per photo type; with it off, no image leaves the warehouse.
Consultation: From the Tracker: by document, by pallet, list between dates and summary by type and model.
Integration: Against Caotic: documents, lines, items, pallets and the item master. No data duplicated anywhere else.

Questions

Do we need to buy new handhelds?: No. It runs on the Android RF handhelds the warehouse already has. If one is very old we check it first, but the normal case is install and go.
Does the operator have to learn another app?: They sign in with their badge or code, scan the label and shoot. The screen tells them which photo is next. Real training takes fifteen minutes.
What if we do not want AI validation?: Leave it off. It is an option per photo type: you can have the AI on the inbound container and not on the rest.
What if the AI gets it wrong?: It does not decide on its own: the photo is stored with the verdict and the reason, and the warehouse sees it in the Tracker. It is built to be cautious — when in doubt it flags the photo rather than passing it — and if it ever does not answer, the photo is stored anyway.
Who can see the photos?: Whoever signs in to the Tracker with permission to see them. At a logistics operator each customer sees only their own goods; the warehouse sees them all.
Do item photos get mixed between documents?: Only if you want them to. Turn the catalogue on and the photo is shared as the item photo; leave it off and it stays on that document — which is what you want for returns and damaged goods.
How many photos can be taken per item?: As many as you set per type, or no limit: the operator closes when done. If the type updates the catalogue, photographing the same item again replaces the previous set of catalogue photos instead of piling up.
What about receipts already closed?: The photos and their index can be rebuilt from the files on the server: each file name carries the reference and the moment it was taken.
Does this work for outbound, not just inbound?: The module works per document type. Inbound and receipts are the usual ones, but loading a truck is configured the same way.

Shall we try it on one of your receipts?

Bring a delivery note, a handheld and any load. In half an hour the capture type is set up, the AI is validating and the photos are in the Tracker.

Request a demo Email info@e-binari.com

+34 93 866 37 30
Travessia, 14, Sant Feliu de Codines
Monday to Friday, office hours