SpecialityVision AI
The cameras and images already in your business, turned into counts, scores, alerts and records you can search. On a phone, a fixed camera, a drone or a satellite pass, running on site or in your cloud. Pixels, decoded.
T25lIGZyYW1l One frame
One frame, six readings
Detection is only the start. The same frame can be segmented, tracked, read, measured and described in plain language, and each reading answers a different question.
model YOLO26
Detect · output
- car0.97
- stop sign0.98
- person0.94
- traffic light0.91
SW4gZGVtYW5k In demand
Where clients point their cameras
Most of the value is in images a business already collects and nobody has time to look at. These are the requests we hear most, and a sample of what comes back.
- 01
Roads and infrastructure
Signs, markings, potholes and street assets inspected from a dashcam, scored and put on a map.
sign RA-1 · legibility 2/5 · replace
- 02
Manufacturing quality
Defects caught on the line at line speed, with the image kept for every reject.
unit 4471 · scratch · reject
- 03
Retail and shelves
Gaps, misplaced products and planogram compliance, from a phone walk or a fixed camera.
aisle 7 · 3 gaps · 92% compliant
- 04
Construction and safety
Helmets, vests and exclusion zones on site, with an alert before an incident, not after it.
zone B · no helmet · supervisor alerted
- 05
Healthcare imaging
Triage and measurement on radiology images, as decision support beside the radiologist.
study 82-114 · flagged for priority read
- 06
Agriculture and land
Trees counted, crops scored and change detected from drone and satellite passes.
plot 12 · 1,284 palms · 3% stressed
- 07
Logistics and yards
Container codes, plates and damage read at the gate, pallets counted in the warehouse.
gate 2 · MSKU 448210 7 · no damage
- 08
Documents and IDs
IDs, forms and receipts read in Arabic and English, with quality checks at capture.
ID · fields extracted · liveness passed
V2hlcmUgaXQgcnVucw== Where it runs
On the device, on site, or in your cloud
Where the model runs decides the latency, the cost and what leaves the building. We choose it per camera, not per project.
A small GPU box on the premises watches many cameras at once. Video stays in the building; events and snapshots go up.
- Leaves the site
- Events and snapshots
- Fits
- Factories, stores, warehouses, any site with many fixed cameras
- NVIDIA Jetson
- DeepStream
- Hailo
- Intel OpenVINO
- TensorRT
RGF0YQ== Data
Night is a different dataset
A model is only as good as the images it learned from. Ours learn from your cameras, at your hours and in your weather, and keep learning after launch.
Frames from your own cameras, from every site and every shift, day and night.
Annotated with tools your team can use too, and checked twice on the hard cases.
Fine-tuned from strong open models rather than trained from zero.
Scored per site, per lighting and per class, so a weak spot shows before launch.
Exported for the hardware it runs on, and quantised where it pays.
Low-confidence frames are flagged and become the next batch to label.
The toolkit
Open models where they are strong, our own fine-tunes where they are not, and the plumbing that gets video in and events out.
- Models
- YOLO26
- RF-DETR
- Segment Anything
- Grounding DINO
- CLIP and SigLIP
- Depth Anything
- Vision-language models
- Reading text
- PaddleOCR
- Tesseract
- Azure Read
- Arabic and English
- Video
- RTSP
- ONVIF
- GStreamer
- FFmpeg
- WebRTC
- Data
- CVAT
- Label Studio
- Roboflow
- FiftyOne
- Synthetic data
- Capture
- iOS and Android apps
- Dashcams
- IP cameras
- Drones
- Satellite imagery
UHJvb2Y= Proof
Every sign, scored at night.

Waysights Road safety, municipalities
AI inspection of road signs for night visibility and legibility, from the Waycam capture app to the Wayboard compliance dashboard. The City of Calgary is the reference deployment.
4x faster inspections
- Waycam: the capture app on the road
- Night visibility and legibility, scored per sign
- Wayboard: the map, the scores and the work orders
- Reference deployment: the City of Calgary
Mounir Remache, Lead Engineer
UHJpdmFjeQ== Privacy
Seeing, responsibly
Cameras point at people. What is masked, what is kept and who decides are designed before the first frame is stored, not argued about after.
Blurred where it is captured
Faces and plates masked on the device or the site box, before anything is stored.
Counting, not identifying
People are detected and counted, not recognised, unless a lawful use requires it.
Kept only as long as needed
Retention set per camera, with events kept longer than footage.
Fair across conditions
Accuracy checked across lighting, weather, camera angle and skin tone.
A person on high-stakes calls
Clinical, safety and enforcement decisions stay with a reviewer.
Within the law
Built for GDPR, the EU AI Act and the Gulf data-protection laws.
Tell us what you want to build
Thirty minutes is usually enough to tell whether there is something here. No deck, no discovery questionnaire.
- DohaQatarHeadquarters. Address to come.
- DubaiUnited Arab EmiratesOffice. Address to come.
- AlgiersAlgeriaOffice. Address to come.
- ParisFranceOffice. Address to come.