Kaya

CNC machine runtime reports from existing factory CCTV. Kaya reads tower light states (green, yellow, red, off) using computer vision. No new hardware. Fully automated.

Try It Now

What Kaya does

Every CNC machine has a tower light. Green means running. Yellow means idle. Red means fault. Kaya watches these lights through existing cameras and turns them into machine activity data. Automatically, continuously.

Reads tower light states

Computer vision models classify each machine's tower light as green, yellow, red, or off. No sensors attached to the machine. No PLC integration. No OPC-UA.

Generates runtime reports

Machine uptime, idle time, and fault duration calculated from tower light data. Delivered through a reporting dashboard powered by Amplify.

Works across CNC brands

Tower lights are universal. Kaya works with Fanuc, Siemens, Heidenhain, Mitsubishi, and any CNC machine with a standard signal tower.

Zero new hardware

Most machine monitoring systems require proprietary sensors, PLC integrations, or edge boxes installed on every machine. Kaya requires none of that.

Uses existing CCTV

Cameras are already mounted on the factory floor. Kaya connects to them directly. No procurement, no installation downtime, no production stoppage.

No sensors or PLCs

No hardware bolted to machines. No PLC integration. No wiring. The camera feed is the only input Kaya needs.

Edge deployment via Greengrass

Computer vision models run at the edge on AWS Greengrass, close to the camera feeds. Low latency, minimal bandwidth to the cloud.

The pipeline

Three SageMaker pipelines handle the full cycle from raw frames to production reports. One deliberate manual step in a fully automated chain.

Auto-annotation pipeline

Frames arrive in S3. AI models generate annotations for tower light states automatically. Every annotation passes through a human review gate before it becomes training data.

Model training pipeline

Reviewed labels feed into SageMaker training. Models improve continuously from real production data, not synthetic benchmarks. New model versions deploy automatically.

Activity reporting pipeline

Machine telemetry and session data flow to DynamoDB. The reporting dashboard reads directly from it. Uptime, idle time, faults, all calculated and available in real time.

Fully automated

From client onboarding to frame ingestion to dashboard visibility. The factory doesn't assign IT staff and doesn't stop production.

Automated onboarding

Signup triggers Greengrass provisioning. Frames start flowing. Models begin training. Reports appear on the dashboard. No manual setup required.

Self-improving models

Production data feeds back into training automatically. Models adapt to new lighting conditions, camera angles, and machine configurations over time.

Multi-facility ready

One deployment serves multiple factories. Each facility's data is isolated. Scale from a single shop floor to a global fleet.

Technology

Deployed entirely on AWS via CDK. Fully reproducible infrastructure. Everything automated from provisioning to monitoring.

SageMaker

Three managed pipelines: auto-annotation, model training, activity reporting. Reproducible experiments with production-grade serving endpoints.

Greengrass

Edge inference runtime on factory-floor devices. Runs computer vision models close to the camera feeds with minimal cloud dependency.

S3 & DynamoDB

Frames, model artifacts, and reports secured in S3, encrypted at rest. DynamoDB stores machine telemetry and session data.

Cognito

Multi-tenant authentication with PKCE flows. Secret management with rotation. Robust IAM policies across all services.

Amplify

Reporting dashboard. Connected through a Lambda/FastAPI API layer to DynamoDB and S3. Real-time visibility into machine activity.

CDK

All infrastructure defined as code. Fully reproducible deployments. Automated from client onboarding to resource provisioning.

Security

Security is fundamental to the design, not an afterthought.

Encryption at rest and in transit

All data in S3 and DynamoDB encrypted using AWS-managed keys. TLS for every connection between services and clients.

Secret rotation

Credentials and API keys rotate automatically. No static secrets stored in configuration or environment variables.

PKCE authentication

Cognito with Proof Key for Code Exchange. Protects against authorization code interception across all client applications.

Role-based access

Four permission levels: system admin, client admin, operator, viewer. Each scoped to specific actions and data boundaries.

Tenant data isolation

Each factory's frames, models, and reports are stored in isolated partitions. No cross-tenant access. One client never sees another client's data.

Audit logging

CloudTrail records every API call. S3 access logs track frame retrieval. Full audit trail for compliance reviews and incident investigation.

Edge device certificates

Greengrass devices authenticate with X.509 certificates and mutual TLS. Each factory device has a unique identity. Revocable at any time.

VPC isolation

Backend services run inside private subnets with no direct internet access. Traffic flows through managed endpoints and security groups.

Data retention policies

Automated S3 lifecycle rules delete raw frames after a defined retention window. No factory footage persists longer than operationally required.

See Kaya in action

We'll walk you through the system with your factory's constraints in mind.