Pre-launch · design-partner pilots open for 2027Synthetic defects · edge inspection · NVIDIA stackApply for the pilot

Platform

One loop from CAD to the reject gate

Viseena is five products on shared data, models and NVIDIA infrastructure. The platform keeps every SKU, model version and verdict connected, so the line gets better with each changeover.

System summary

Products
5 on one data loop
Edge
Jetson Orin or IGX
Training
DGX/HGX or cloud GPUs
Deployment
Edge plus cloud
Status
Pre-launch
Architecture

Six layers, each with one job

LayerWhat it doesRuns onProduct
Synthetic-data studioIngests CAD, materials and a defect taxonomy; renders labeled datasetsDGX/HGX or cloud GPUsSynvex
Training pipelineFoundation model pretraining, per-SKU adapters, sim-to-real alignmentDGX/HGX or cloud GPUsCoreon
Edge inferenceMulti-camera detection, segmentation and anomaly flagging at line speedJetson Orin or IGX at the stationVeridex
Reasoning layerDefect reports, triage and root-cause analysis over verdicts and telemetryGPU inference, cloud or on-premisesPilora
Digital twinVirtual camera and lighting placement and coverage checksRTX workstation or DGXTwinix
Control planeModel registry, fleet deployment, audit trail, access controlViseena Cloud or your cloud (planned)Platform
The flywheel

Every line makes the next model better

Shared defect taxonomies and the base model improve with every SKU, while your images and adapters stay yours.

Deployment

Run it where your plant needs it

  1. 01

    Edge plus Viseena Cloud

    Inspection runs on the edge node. Synthetic generation and training run in Viseena Cloud on NVIDIA GPUs.

    Default
  2. 02

    Edge plus your cloud

    Training and the control plane run in your own cloud account, with the same edge runtime.

    Planned
  3. 03

    Air-gapped plant

    Everything, including the Pilora copilot, runs on-premises on IGX and DGX hardware.

    Planned
Control plane

Every model, every station, one registry

The control plane versions each SKU model, records which version ran on which station and when, and rolls out or rolls back updates across lines and plants.

  • Per-SKU and per-station model versions with rollback
  • Staged rollouts: one camera, one line, then the plant
  • Audit trail linking each verdict to its model and data
FleetExample

Plant 1 model rollout

Line A · ST-2
Adapter v7
Line A · ST-3
Adapter v7
Line B · ST-1
Adapter v6
Line B · ST-2
Staged v7
Line A100%
Line B50%
Line C0%
Example. Values are illustrative.
Security

Built for plant IT and OT

  1. 01

    Frames stay on the line

    Edge nodes process frames locally. Sending frames to training is opt-in per station.

    Data
  2. 02

    Role-based access

    Operators, engineers and administrators get separate permissions, with single sign-on (planned).

    Access
  3. 03

    OT-friendly by design

    Edge nodes make outbound connections only, and can run with no internet in air-gapped mode (planned).

    Network
  4. 04

    Signed records

    Model versions, approvals and verdict exports are logged with who, what and when.

    Audit
  5. 05

    Supported stack

    Built on NVIDIA AI Enterprise components with tracked versions.

    Supply chain
  6. 06

    Independent review

    A third-party penetration test and an ISO/IEC 27001 programme are planned before general availability.

    Assurance
Next step

See the platform on one of your parts

Send a CAD file and a defect list. We will show you a synthetic run for it.