Computer vision development
Computer vision lets software understand images and video: detect people, vehicles or objects, spot defects, read labels and trigger actions when something happens. Athryx Tech builds computer vision systems that run on real camera feeds and connect to the tools your team already uses, for clients in Romania and across Europe.
We built AI vision pipelines for NEI Security's surveillance workflows, turning camera streams into events operators can act on. Through Athryx, our robotics venture, we also work on inspection robotics for non-destructive testing (NDT) and critical infrastructure, where vision and sensor data replace slow manual checks.
A vision model is only one part of the system. It needs reliable video input, hardware that can run it at the edge or in the cloud, alerts that reach the right person and a way to review mistakes. We build all of it, and we measure success in fewer missed events and less manual monitoring.
Problems we solve
- Operators watching camera walls for events that rarely happen
- Manual visual inspections that are slow, costly or dangerous
- Video recorded for weeks and never analysed
- Vision pilots that work on sample footage but fail on real cameras
Who this is for
- Security and facilities teams with existing camera networks
- Manufacturers that need visual quality inspection
- Infrastructure and industrial operators running inspections
- Product teams adding image or video understanding to their software
What we deliver
- Object, person and vehicle detection on live video
- Event detection and alerting for security workflows
- Visual inspection and defect detection
- Edge deployment on GPUs close to the cameras
- Integration with video management, ticketing and operations tools
- Review dashboards to check detections and improve models
Technologies & stack
- PyTorch and ONNX for model training and export
- NVIDIA GPUs and CUDA for edge and cloud inference
- Docker and Kubernetes for deployment
- Python and TypeScript services and APIs
- RTSP camera stream ingestion and video pipelines
How we work
- 01Review cameras, environment and the events that matter
- 02Prototype on your real footage, not sample data
- 03Deploy to edge or cloud with monitoring and alerts
- 04Review false positives with your team and retrain
Selected reference
NEI Security: AI vision pipelines for surveillance workflows. Athryx: inspection robotics and physical AI for NDT and critical infrastructure, currently preparing for Pre-Seed.
Frequently asked questions
What is computer vision used for in business?
Security monitoring, quality inspection, counting and tracking, safety compliance and reading documents or labels: anywhere people currently watch video or check items by eye.
Can computer vision work with our existing cameras?
Usually, yes. Most IP cameras provide a video stream we can process. We check resolution, angles and lighting first, because they decide what can be detected reliably.
Edge or cloud processing?
Edge when you need low latency, have limited bandwidth or video must stay on site; cloud when you need scale or central processing. Many systems use both.
Do you work with clients outside Romania?
Yes. We work in English with clients across Europe. Model development happens remotely, before an on-site rollout.
Related services
Have cameras but no insight?
Tell us what you need to detect and where. We will tell you what is realistic with your current cameras and what a pilot would look like.
Get in touch