AI Visual Inspection

The Sentinel Platform

A complete machine vision platform for building, training, and deploying inspection models

How It Works

Import

Browse images from local storage in-place

Create

Define ROIs & choose model type

Train

GPU-accelerated with live metrics

Inspect

Multi-model batch inspection

Improve

Review, dispute & retrain

4 AI Model Types

Click any model to learn more

Classifier

Supervised

Pass/fail classification using CNN architectures

Train with labeled good and defective images. Supports ResNet, EfficientNet, and MobileNet backbones with configurable hyperparameters.

Best for: Simple pass/fail, multi-class defect categorization

EfficientAD

State of the Art

Anomaly detection using only good images

Industry-leading anomaly detection that learns what "normal" looks like. Produces pixel-level heatmaps showing exact defect locations. Requires only good samples for training.

Best for: Surface defects, unknown defect types, minimal training data

Autoencoder

Unsupervised

Detect anomalies via reconstruction error

Trains on normal images only — any deviation from the learned pattern triggers an alert. Uses a bottleneck architecture to learn compressed representations.

Best for: Texture anomalies, manufacturing consistency

YOLOv8 Detection

Object Detection

Locate and classify defects with bounding boxes

Identify multiple defect types and their exact positions in a single pass. Supports nano to large model sizes for the right speed/accuracy tradeoff.

Best for: Defect localization, multi-defect scenes, position-based rules

Dashboard Features

Templates

Templates

Define what to inspect by drawing regions of interest (ROIs) on a golden reference image. Group identical regions across templates to share datasets and models.

  • Visual ROI editor with drag & resize
  • SAM-powered auto-segmentation
  • Region groups across templates
  • ORB-based image alignment

API & Integration

Built API-first for seamless production integration

REST API

Full-featured API with Swagger docs. Every capability is accessible programmatically.

MES / SCADA

Trigger inspections, receive results, and automate quality gates in your MES.

PLC / TCP/IP

Direct PLC communication for in-line inspection trigger workflows.

Watch Folders

Auto-inspect new images as they arrive in configured directories.

Closed-Loop Quality

Traditional systems are static. Sentinel learns from every inspection through its feedback pipeline.

1
Inspect images with trained models
2
Quality engineers review results
3
Disputed results become training data
4
Models retrain with corrected labels
5
New version deployed — accuracy improves

See Sentinel in Action

Schedule a personalized demo to see how Sentinel fits your production environment.

Request a Demo