Computer Vision Research & Product Studio
We build vision products that read the physical world
Ebsar is a computer vision studio building and shipping our own products for automotive, insurance, and retail.
Live in production
Our Products
We build and run our own computer vision products. Three are live in production today, with more in development.
Automotive · Parking & Traffic
ANPR.software
Real-time automatic number plate recognition, running in production in the UAE. Detects plates, reads them via OCR, and classifies vehicle make, model, and color from a single frame.
- — Real-time plate detection and OCR
- — Car brand, model, and color identification
- — REST API with 1s response
Insurance · Fleet & Rental
DamageDetection.software
AI vehicle damage assessment via API. Detects dents, scratches, cracks, broken glass, rust, and missing parts from vehicle photos — built for insurance claims, rental check-ins, and fleet inspections.
- — 98%+ detection accuracy, sub-5-second analysis
- — Severity grading mapped to specific parts
- — Free pilot — 100 analyses, no credit card
Automotive · Multi-Country Plates
LPR.software
License plate recognition tuned for plate formats beyond the GCC. A shared recognition core that handles Latin, Arabic, and Cyrillic scripts across Central Asian and European plate standards.
- — Multi-country, multi-script plate formats
- — Deployable on edge devices via ONNX
- — One integration, many regions
Retail · In-Store Analytics
Retail Vision
In-store analytics from existing CCTV — counting visitors, tracking dwell time, detecting growing queues, and monitoring shelf stock, processed on-device for privacy. In active development.
- — Foot traffic and dwell-time analytics
- — Queue detection and shelf-stock monitoring
- — On-device processing, privacy by design
Flagship Product · Live in the UAE
ANPR.software — Real-Time License Plate Recognition in Production
Our flagship product runs a full ML pipeline that processes vehicle images in real time. It detects license plates, reads plate numbers via OCR, classifies country of origin, and identifies car brand and model — all from a single photograph. The same research core powers LPR.software and DamageDetection.software.
1s
Inference per image
97%+
Plate read accuracy
Vehicle Detection
DetectionLocalize all vehicles in frame using real-time object detection.
License Plate Localization
DetectionCrop and align plate region from each detected vehicle.
Plate OCR
RecognitionRead characters from plate image — Latin and Arabic script.
Car Brand & Model Classification
ClassificationIdentify make, model, and color from vehicle photograph.
Where We Focus
We go deep in three markets where a camera and a fast, accurate decision are worth real money. Each has a product behind it.
01
Automotive
Plate recognition, vehicle identification, and access control for parking operators, traffic authorities, and smart-city infrastructure.
ANPR.software · LPR.software
02
Insurance
Instant photo-based damage assessment for claims, rental check-in/out, and fleet inspections — cutting manual review from days to seconds.
DamageDetection.software
03
Retail
In-store analytics from existing cameras — footfall, dwell time, queues, and shelf availability — turning CCTV into operational insight.
Retail Vision · in development
From Research to Product
How an idea becomes a live product. The same pipeline runs behind every product we ship.
01
Research
We pick a real problem in automotive, insurance, or retail and study the data, edge cases, and state of the art before writing a line of training code.
02
Train & Benchmark
We build datasets and train models against measurable targets — accuracy, latency, robustness — iterating until they beat the bar in the wild, not just on paper.
03
Productize
Winning models become a product: an API, a dashboard, and integrations, optimized to run in real time on cloud or edge.
04
Operate & Improve
We run the product in production — monitoring, retraining on new data, and detecting drift — so it gets sharper the longer it lives.
How We Operate
A product studio at heart. We own what we build end to end — the research, the models, the APIs, and the roadmap. Every system we run is trained, tested, and maintained by the same team.
Products, Not Projects
We build and operate our own software. Three products are live in production and generating real usage — not prototypes on a shelf.
Research-Driven
A shared research core — detection, OCR, fine-grained classification — powers every product. Advances in one lift them all.
Vertical Focus
We go deep in automotive, insurance, and retail rather than wide across everything. Domain data and hard edge cases are our moat.
Production-Proven
Our ANPR product runs in production in the UAE, processing real traffic at 1s per image. Deployed, monitored, and maintained.
Fast to Pilot
From first conversation to a working pilot on your data in weeks, not quarters. Reusable architectures and deployment patterns let us move quickly.
Deploy Anywhere
Cloud API, on-premise, or edge via ONNX. Data stays where it belongs — no vendor lock-in, no data leaving the network.
Try our products — or build with us
Run a free pilot on one of our live products, or talk to us about partnerships, deployments, and what we're building next in automotive, insurance, and retail.
Founder
Abzal Assembekov
Abzal is the founder of Ebsar, a UAE-based computer vision studio building applied AI products for automotive, insurance, and retail. Over 15 years he has founded multiple software companies and led engineering teams shipping production software worldwide.
At 20, his startup Propertyin360 was named by Forbes Middle East one of the 50 UAE Startups to Watch. He later built ContactInBio to 100,000+ users and one million monthly visitors before its acquisition by Taplink in 2021. Today he leads Ebsar's product and engineering, developing ANPR.software, DamageDetection.software, and LPR.software.
More about Ebsar →