home / use cases / documents and vision / photo-based-inspection

Field photos become records, not backlog

Drivers and technicians send photos from the field — delivery notes, meter dials, damage, site checks. A vision model running on your own box reads each one and writes a structured record into your database, flags damage, and asks for a retake when a shot is unusable. Images never leave your network.

Watch the demo

starts when: A driver or technician sends photos in from the field

The problem

Field work produces photos by the hundred: the signed delivery note on the pallet, the meter before and after, the scratch that was already there. And then the photos produce nothing. They sit in a messaging thread or a camera roll until someone in the office retypes what they show — or, more often, until a dispute arrives and nobody can find the one shot that would have settled it.

The retyping is classic hidden admin — the sort of repetitive transfer that eats most of an employee’s working time without ever appearing on a job sheet. And the photos themselves are awkward to push through cloud tools: they show customer sites, plates, people, the inside of buildings you were trusted to enter.

How your Aivell does it

  1. Photos come in the way they already do. Drivers email them from the road; a field app can push them over a webhook. Nobody’s routine changes.
  2. Each shot is read on the box. The onboard vision model reads the handwritten note number, the dial position, the label — and sees the dent. No image is uploaded anywhere to be understood.
  3. The photo finds its job. Note numbers, plates and timestamps tie each shot to the delivery or work order it belongs to, so the record lands in context instead of in a folder.
  4. A structured record is written. Job, reading, condition, timestamp — inserted into your database, one row per event. Insert-only: existing records are protected by a guardrail.
  5. Exceptions go to people, with the picture attached. Visible damage opens a damage record and pings the person you named. An implausible meter jump waits for approval. A useless photo triggers a retake request that says exactly what to reshoot.

By evening there is a mailed recap listing everything that came in and what became of it. The judgment calls — what counts as damage, how far a reading may drift, who gets woken up — are lines in a chore, written in plain language, changed in seconds.

The chore, as you'd write it.

No flowcharts, no code — a chore is just a message to your Aivell, in your own words. It shapes it into a solid, guarded procedure and follows it to the letter. This one:

Giulia Aivell · log field photos ● active

accept photos mailed by drivers or pushed from the field app

read each one — delivery note numbers, meter dials, visible damage

match the photo to the job or delivery it belongs to

write a record — job, reading, condition, timestamp

damage in the shot? open a damage record and send Davide the photo

meter reading far off last month's? ask before saving

blurry or wrong subject — reply asking for a retake, and say what's missing

at day's end, mail Davide the list — every photo, every record written

You
Got it — I'll start right away, and check with you whenever something needs a human.
Giulia Aivell

The tools it uses.

Connected with single-click passkeys — no copied tokens, no OAuth hell.

Email receives photos mailed from the road
Webhooks accepts uploads from your field app
Documents & Vision reads dials, labels and damage onboard
Database stores one structured record per photo

Sensible guardrails.

Every tool can be limited to exactly what this chore needs. For this one, you might set:

Database May insert new inspection records. Updating or deleting existing rows is blocked.
Email Replies only to the field colleague who sent the photos; no external addresses.

Questions, answered.

What can it actually read out of a photo? +

The things your back office currently squints at — handwritten delivery notes, meter dials and digital displays, serial-number labels, pallet counts, and the state of what's in frame, like a dented panel or a cracked windscreen. Each photo becomes fields in a record, not a file in a folder.

The photos show our customers' premises. Are they uploaded to some vision service? +

No. The vision model lives on the box in your office, and the photos are read there — behind your firewall. Pictures of a client's warehouse, gate codes on a wall, or a person caught in frame stay inside your network, full stop.

What if the photo is too blurry to trust? +

The chore says ask, so it asks — a reply goes back to the driver naming exactly what is missing ("the dial is out of focus, reshoot closer"). Nothing gets saved on a guess, and readings that jump implausibly from last month wait for a human yes.

Do drivers need a new app? +

No. Email works from any phone today. If you already run a field app, it can push photos straight to the box over a webhook — both roads end in the same record.

Related.

What is an Aivell?

An Aivell is an on-premise AI colleague: a small box that plugs into your network, runs its own AI onboard and takes the repetitive work off your desk. Prompts, documents, data — nothing ever leaves your office. Unplug the internet: it keeps working.

You describe each task as a chore, in plain language. Your Aivell turns it into a solid, guarded procedure and runs it in the background — with guardrails on every tool, a complete audit trail, and approvals in your hand when you want them. One-time setup, fixed monthly fee. No tokens, no overages.

Watch the demo

onboard inference · air-gapped · fixed monthly fee