Stages 3–4
Whoever trains the model deploys it.
The researcher puts their own model on the client’s server and watches the first week themselves. Which is why “it worked in the lab” is not an answer anyone here can give.
The team
The team that turns research into production. Experience earned in Korea, now at work on floors in Uzbekistan.
Bayonic is built by engineers who trained in South Korea, the United Kingdom and Tashkent — people who took AI into production in automotive, factory, port and city programmes. That experience now has one job: turning the cameras you already own into a measuring tool.
- South Korea
- United Kingdom
- Tashkent
- 1st placeICCV · instance segmentation
- 70+research papers · IEEE, ICCV, CVPR
- 12patents
- 10+ yearsof production AI
- CES2024 and 2025 — on stage two years running
- Grand prizeKorea Transportation Safety Authority
BayonicTashkent
Every degree, paper and award on this page belongs to a named member of the team. Clients in Korea are not named — that is a legal obligation; the work is described at sector level only.
Leadership and the senior engineers.
Each of them has taken a model into production rather than into a paper — in automotive, factory and city programmes. Below, briefly: who they are, where they trained and what they shipped.
- Projects delivered in Korea
- 20+automotive, electronics, port and city programmes
- Trained in
- 3South Korea · United Kingdom · Uzbekistan
Shohruh Rahmatov
Founder & CEO
ICCV — 1st placeBuilt real-time vision systems that run in global industry.
- Education
- M.Sc — Kumoh NITSouth Korea
- Experience
- Global automotive & safety programmes · 12+ years
Jahongir Yunusov
Chief Technology Officer
ICCV — 1st placeResearch engineer who took multi-camera vision into production.
- Education
- M.Sc — Kumoh NITSouth Korea
- Experience
- Automotive, port and factory programmes · 12+ years
Suhrob Jurayev
Strategy Director
Finance and corporate strategy — macroeconomics and investment modelling.
- Education
- M.Sc — Swansea UniversityUnited Kingdom
- Experience
- Chief macroeconomist at a national bank · 10+ years
Ikrom Ibrohimov
Lead Backend Engineer
The platform under the load: APIs, Kubernetes, gRPC and MLOps pipelines.
- Education
- B.Sc — INHA UniversityTashkent
- Experience
- High-load production systems · 6+ years
Salohiddin O'roqov
Senior AI Researcher
Research in segmentation and model compression.
- Education
- PhD candidate — Kumoh NITSouth Korea
- Experience
- A CVPR-level research lab
Universities and the lab
- KAIST
- Sungkyunkwan
- Kumoh (KIT)
- Kyung Hee
- Gachon
- CVPR Lab
- NVIDIA Inceptionmember
Korean employers and clients are not named — that is a legal obligation. The sector, the result and the person are public.
The team is larger than this: the page publishes the leadership and the senior engineers. The rest of the profiles follow when they are ready.
The model is born on your floor and measured on the same floor.
Work here is one closed loop: the data comes off your cameras, the model runs on your server, the result is measured in your console — and that measurement sets the next version. Below is that loop, exactly as it runs.
- 01
Your floor
The cameras, shifts and till you already have — not an open dataset.
- 02
Training
The model is fitted to this room, this lighting and this process.
- 03
Your server
An on-premise deployment, done by the engineer who trained the model.
- 04
The first week
That same engineer watches the output daily and tunes the thresholds.
- 05
Measurement
The figure is read in your console — the only figure we ever publish.
The measurement sets the next version — the loop closes
The three rules that hold the loop together
Stage 5
The number comes off the floor, not off a benchmark.
Every result this site publishes was measured in the client’s own console. What a model scored on its test set is internal work, not a claim.
Stages 1–3
Data never leaves the client’s server.
Neither the video nor the staff records go anywhere — the model runs on the client’s own hardware. What is stored and what is not has its own page.
The security pageWant this team on your floor?
Which cameras are already there and what you want measured — that is where the conversation starts. What this team has actually built is written up on the measured-results page.