01
Expensive cloud inference
Per-photo GPU inference and storage costs grow linearly with every store visit — margins shrink as you scale.
SKU 98%
On-device · Android · SDK
On-device shelf recognition SDK for trade marketing & SFA platforms
Detect SKUs and calculate retail execution KPIs directly on Android devices — without cloud processing, GPU infrastructure or internet dependency.
92%
OSA
47
Facings
3
Price issues
Built on retail execution experience across global CPG programs
CPG brand Retail chain Field agency SFA platform Distributor The problem
Cloud shelf recognition is expensive to scale
Every visit, photo and recognition request creates infrastructure cost, latency and a dependency on store internet. The economics break the moment you roll out to real field teams.
02
GPU & server infrastructure
You maintain inference clusters, queues and autoscaling — a CV ops team just to keep recognition running.
03
Slow processing & sync
Merchandisers wait for uploads and results. Multiply that by hundreds of visits a day and time evaporates.
04
Internet dependency in stores
Back rooms and basements kill connectivity. No signal means no recognition — exactly where it's needed.
05
Manual annotation bottleneck
New categories and retailers each need costly hand-labeled datasets before a model can even be trained.
06
Complex rollout
Pushing recognition to agencies and thousands of merchandiser devices becomes an operational project of its own.
How Jabex works
From a shelf photo to structured KPIs — on the device
The merchandiser never changes their workflow. Jabex runs inside your existing app and only the structured result leaves the phone.
1
Take photo
In your existing SFA app, as today.
2
On-device recognition
Compact ML model runs locally on Android.
3
SKU detection
Products, facings and price tags identified.
4
KPI calculation
OSA, share & compliance computed locally.
5
Structured sync
Only result data is sent to your server.
6
Dashboard / SFA
Clean KPIs land in your existing system.
On-shelf availability (OSA)
SKU presence & facings
Price tag compliance
Secondary placement
Shelf share & planogram checks
Why on-device
Move image processing to the merchandiser's phone
A compact 10–30 MB model processes shelf photos locally. No upload, no GPU bill, no waiting on store Wi-Fi.
No cloud processing cost
Recognition happens on the device, so there's no per-photo inference bill to absorb as you scale.
$0 / inference
No GPU infrastructure
Skip the inference clusters and CV ops entirely. Nothing to provision, autoscale or keep online.
0 servers to run
Works offline
Back rooms, basements, rural stores — recognition runs with zero connectivity, every time.
100% offline-capable
Lightweight model
Just 10–30 MB ships to the device — small enough to bundle and update painlessly.
10–30 MB on disk
Fast deployment
Drop the SDK into an existing Android app and start recognizing — no platform rebuild required.
SDK-level integration
Model updates without app release
The ML model is a downloadable file — push new categories or retailers without an app store cycle.
Hot model swaps
SDK integration
Jabex adds the recognition layer. You keep everything else.
Delivered as an Android SDK that embeds into your existing SFA, field force or merchandiser app. Your UI, workflows and backend stay exactly as they are.
Your layer
Existing SFA app
The merchandiser interface, routes and workflows your teams already use.
- UI & UX unchanged
- Visit & route logic
- Photo capture
Jabex SDK
On-device recognition
The compact ML model detects SKUs and computes execution KPIs locally.
- SKU & facing detection
- KPI engine
- Downloadable model
Your layer
Structured results API
Only clean, structured recognition data syncs to the backend you own.
- JSON result payloads
- Your database
- Your dashboards
photo in → on-device inference → structured JSON out
Android SDK
Fast implementation
Existing workflows stay intact
Backend remains yours
Faster model training
No expensive manual annotation projects
Traditional shelf recognition needs costly hand-labeling before a model exists. Jabex uses Vision LLM auto-labeling on real retailer shelf photos to prepare datasets faster — so launching a new category, retailer or portfolio stops being a data project.
New categories
New retailers
New portfolios
01
Retailer shelf photos
Real captures from the field
02
Vision LLM auto-labeling
Datasets prepared without manual annotation
03
Model training
Compact on-device model produced
04
Downloadable update
Shipped to devices without an app release
Design partner program
Looking for early partners
We're onboarding a small group of trade marketing agencies, SFA platforms and CPG teams to validate Jabex on real shelves.
Pilot launch
A scoped recognition pilot on your categories and stores.
SDK integration workshop
Hands-on session to embed Jabex into your app.
Category validation
We tune and validate recognition for your portfolio.
Agency / SFA partnership
Embed shelf recognition as a new premium module.
FAQ
Questions enterprise buyers ask
Yes. Recognition runs entirely on the Android device using a local model, so it works with no connectivity at all. Only the structured result syncs when a connection is available.
Jabex targets the broad range of modern Android devices used by field teams. Exact minimum versions and device requirements are confirmed during SDK scoping for your fleet.
The ML model is a downloadable file (10–30 MB). New categories, retailers or portfolios ship as a model update — no app store release or full app update required.
Yes — Jabex is delivered as an SDK that embeds into your existing app. Your UI, workflows and backend stay intact; Jabex adds the recognition layer and returns structured results.
No. There is no cloud image processing and no GPU infrastructure to run. Inference happens on the merchandiser's device, so there are no per-photo server costs.
Because Jabex slots in at the SDK level without changing your workflows, integration is fast. We scope a concrete timeline with you in the SDK integration workshop.
Get in touch
Turn shelf photos into retail execution data
Tell us what you're trying to solve and we'll discuss SDK integration or a pilot project.
