Step 01
Digital trigger
A delivered status from Amazon or Robu.in, or a courier SMS or WhatsApp message, starts the search.
All case studies
Case Study · Workplace · Chain of Custody
Automated Parcel Tracking
Tracking ends at the front door. Korv Glanz picks it up from there: it links each delivery to the camera footage and records who received the parcel and exactly where they left it.
At a glance
Parcel tracking inside the building
Runs on one server next to the existing NVR
Triggers
Amazon and Robu.in order status, plus courier SMS and WhatsApp messages
Vision
Person and parcel detection, with re-identification across cameras
Location
A vision-language model describes where the parcel was placed
Runs on
One on-site server, next to the existing NVR
01 / The challenge
Parcels go missing inside the building
01
Misplaced custody
Whoever is nearest the door signs for the parcel and leaves it on a random desk, workbench or shared area.
02
Blind spots
The person who ordered it can't find it, assumes it never arrived, and the item gets ordered again.
03
Hands-off
Sign-in sheets and mailroom staff don't suit lean teams. The fix had to need no change in anyone's habits.
Where Deliveries Come From
Amazon, Robu.in and every courier message
The search starts from a real delivery, not from round-the-clock watching. Orders are read from the shops the team buys from, and anything else is caught from courier messages.
Amazon
Orders pageAmazon offers no order API for buyer accounts, so a headless browser reads the delivered status and time from Returns & Orders.
Robu.in
Order APIOrder status and the exact delivered time come straight from Robu.in, checked around the expected delivery date.
WhatsApp & SMS
Courier messagesFor local couriers, DHL or Blue Dart, messages like “Delivered” or an OTP confirmation start the search too.
WhatsApp · delivery message
Amazon Delivery Update
Your order has been successfully delivered!
- Product
- Logitech M185 Wireless Optical Mouse
- Delivery date
- 16 October 2026
- Delivery time
- 4:07 PM IST
- Status
- Delivered
Dashboard · parcel handling
- Who
- Employee 1432
- Kept where
- C129, on the seat of an orange chair
- Date
- 2026-10-16
- Time
- 04:06:51 PM – 04:07:21 PM
A real delivery from our own office, from the message that started the search to the answer on the dashboard
02 / How it works
From a delivery message to a sentence that says where
What happens after a delivery is confirmed, from the message to the dashboard.
Step 02
Fetch footage
The system pulls the NVR recordings from around the delivery time.
Step 03
Detect and track
People and parcels are detected; re-identification keeps the same person's identity across cameras.
Step 04
Drop-off
The drop is logged when the parcel stays still and the person who carried it walks away.
Step 05
Describe
A vision-language model looks at the final frame and writes where the parcel is, in plain words.
Step 06
Dashboard
Who received it, where it is and when, with a proof clip and snapshots.
Example answer
“C129, on the seat of an orange chair.”
See It Run
One delivery, start to finish
Recorded in our own office. The Amazon message arrives, both cameras are searched, the person carrying the parcel is identified, and the dashboard says where it was left.
Getting Custody Right
Walking past a parcel isn't the same as carrying it
Early tests credited parcels to whoever happened to be nearby. These rules fixed that, so the name on the record is the person who actually handled it.
01
Possession, not proximity
The parcel counts as carried only when its centre stays inside a person’s box for a sustained run of frames. Someone walking by in the background isn’t credited.
02
A clear drop-off moment
The drop is logged only when the parcel stays still and the person who carried it walks away, so a parcel set down for a second isn’t mistaken for its final place.
03
The same person on both cameras
Re-identification matches clothing, colour, posture and build, so the carrier keeps one identity as they cross from one camera to the other or step behind a pillar.
04
A sharp final frame
For the description, the system picks the clearest unblurred frame of the parcel where it was left, so the location is written from what is actually visible.
On Every Delivery Record
One card answers what, who, where and when
Anyone in the office can open the dashboard and find a parcel without asking around or scrubbing through footage.
- 01Delivery details: product, brand, order and delivery dates and status, taken from the order or message
- 02Who received it: the employee matched on camera
- 03Where it was kept: a plain sentence, such as a desk number and the chair it sits on
- 04When: the date and the time window of the hand-off
- 05Evidence: a snapshot of the drop-off with the person and parcel boxed
Built To Stay Small
It Only Looks When a Delivery Is Confirmed.
There is no round-the-clock video analysis. A delivery message starts a search of the footage around that moment, so everything runs on one server next to the existing NVR.
03 / Outcomes
What changed
- 01No more re-ordering items that were already on site
- 02No time lost searching the office or asking around
- 03A documented chain of custody for every delivery
The system doesn't watch footage around the clock. It only looks when a delivery is confirmed, which keeps the compute small enough to run everything on one on-site server.
Runs On Site
Footage and identities stay in the building
Everything runs on one on-site server. Nothing is sent to the cloud.
| Area | How it is handled |
|---|---|
| Compute | One on-site GPU server runs detection, re-identification and the vision-language model together |
| Cameras | Two IP cameras with overlapping views of the entrance and the main desks, recorded by the existing NVR |
| Footage | Only the few minutes around each confirmed delivery are pulled from the NVR |
| Identity | People are matched against an internal gallery of employee appearance kept on the server |
| Dashboard | Served over the office network, with the record, snapshot and proof clip for each delivery |
Limits and What’s Next
Where it stops today, and what we’re building
Current limits
TodayBlind spots
A parcel put in a drawer, a locker or under a table can’t be described, because the description comes from what the camera sees.
Very small deliveries
Flat envelopes and thin mailers can blend in with paperwork and go undetected.
Missed triggers
If a shop changes its order page and no courier message arrives, the search never starts.
Recording must be on
If the NVR is offline when a delivery lands, there is no footage to search.
On the roadmap
NextAn envelope detector
A small second model trained only on envelopes and flat mailers.
A wider fallback search
If a delivery is confirmed but no parcel is seen, the search widens and flags possible matches for a person to check.
Self-healing order readers
Readers that adapt when a shop’s order page changes, so triggers keep coming.
More cameras
For larger buildings, the work spreads across several edge servers.
See It On Your Cameras
Ready to stop losing parcels?
See how Korv Glanz follows deliveries on your own cameras, and where it finds them.
- ✓Personalized 1-on-1 walkthrough
- ✓A review of footage from your building
- ✓Technical integration assessment
Delivery triggers. Camera re-identification. A sentence that says where.
Request a Live Demo
Let's look at your building.
Tell us a little about your building and our team will get in touch.
