.jpg){kind=link}
NVIDIA Jetson
On-site inference hardware for supported camera streams.
Leak Sight AI Architecture
Review video on site. Move only what policy allows.
Leak Sight AI evaluates supported camera streams near the source and synchronizes incident data or approved clips only when policy allows.
Facility boundary
Supported inputsRGB · Thermal · Sensors
On-site inferenceDetect · Compare · Verify
Policy-controlled outputIncident metadata · Approved clips
Edge & Data Flow
Sensitive video can remain inside the facility.
Supported RGB and thermal streams are tied to known assets and monitored zones. The edge process detects changes, compares them over time, and creates an incident record for operator review.
- Event-only cloud synchronization where policy permits
- Encrypted transfer for approved metadata and clips
- Configurable video and incident retention
- Role-based access controls and audit logs
Facility Boundary
Supported InputsRGB · Thermal · Sensors
On-Site InferenceDetect · Compare · Verify
Policy-Controlled OutputIncident Record & EvidenceEvent metadata or approved clips
Operator ReviewMaintenance Export
Facility BoundaryDense pipework makes monitored zones and asset mapping essential.
Each camera view should resolve to a known asset and region of interest. Edge processing keeps the raw visual stream close to that operating context.
Edge Technology Stack
A practical path for accelerated edge inference.
Leak Sight AI supports NVIDIA Jetson, TensorRT, and DeepStream for accelerated edge inference. Their mention does not imply a partnership, certification, or endorsement.
TensorRT
Optimization layer for efficient model execution at the edge.
DeepStream
Pipeline for supported multi-stream video processing.
Data Strategy
Improve from controlled evidence, not unrestricted collection.
Customer video can expose confidential processes. Model development and calibration should use defined purposes, approved retention, and traceable access.
- 01Site calibration
Establish monitored zones, operating baselines, and known non-leak conditions for each facility.
- 02Rare-condition coverage
Maintain documented coverage for difficult weather, occlusion, exhaust, washdown, and other low-frequency conditions.
- 03Operator feedback
Use confirmed, dismissed, and monitored outcomes to refine facility calibration.
Pilot Qualification
Map the deployment to your network and retention policy.
A pilot should define camera support, edge hardware, event synchronization, access, retention, and maintenance-system handoff before installation.