Computer vision that guides people through real work.
A headset camera streams what the wearer sees to an inference server. Detections come back as world-anchored overlays, step by step, with a remote expert able to draw directly into the trainee’s view. Running on Meta Quest and Magic Leap 2. Offered jointly by Ethar, Inc. and ZenithMoon Studios: Ethar owns the product and leads customer engagements, and ZenithMoon Studios designed and built the platform.
What we do
Vision pipelines to headsets and edge devices
WebRTC video from the device to a vision inference server running interchangeable detector families (YOLO, OWLv2, GroundingDINO, Florence-2 and body pose), meaning fixed-class, open-vocabulary and pose models can be swapped without changing the client. Results render in place.
Fits: Offered jointly with Ethar, Inc. Engagements can start with either partner.Demonstration: AI Vision Training Platform →Remote assistance with expert annotation
A console where an expert sees the trainee’s view, talks to them and annotates directly into their headset.
Demonstration: AI Vision Training Platform →Recognition platform design
Requirements, architecture and test strategy for hosted recognition services, choosing established tooling over bespoke builds: experiment tracking, model serving, annotation, dataset curation and data versioning (MLflow, Triton, CVAT, FiftyOne, DVC).
Adaptive learning platform
Delivered as a complete adaptive learning platform, integrating real-time feedback, fine-tuning, and evaluation gates, with a home-lab-first approach that defers cloud spend until load justifies it.
What this has produced
Told by organisation type and date. Client and product names are withheld. Business partners are named.
Guidance that appears where the work is
A vision AI training platform running five interchangeable detector families, with a step-by-step training engine and a remote-assistant console. Offered jointly by Ethar, Inc. and ZenithMoon Studios: Ethar owns the product and leads customer engagements, and ZenithMoon Studios designed and built the platform.
Meta Quest, Magic Leap 2, experimental WebXR clientDemonstration: AI Vision Training Platform →A hosted recognition service from two prototypes
Product requirements, architecture and test strategy that merge two prototypes into one product, after a market study showed the niche vacated by discontinued cloud products.
Established MLOps tooling over bespoke buildsSee it running.
Working systems that show this practice in use. Capabilities and results are described; how they are built is not.
The rest of the practice
Tell us about the estate, the device or the problem.
Email a few lines about what you are trying to do. We will reply with what we would look at first.