Halcyon Arc / Technology
Halcyon Arc technology

One secure foundation, built cloud-native on AWS and Google Cloud.

Every Halcyon Arc product runs on a shared foundation for identity, encryption, model access and audit, with our AI Safety and AI Governance platforms built in. We design for Canadian health-data residency, least privilege and portability across both clouds.

Architecture principles

Why multi-cloud, and why cloud-native

  • Canadian residency. Production and health data are hosted in Canadian regions: AWS Canada (Central) and Canada West (Calgary); Google Cloud Montréal and Toronto.
  • Best model for the job. We evaluate and route between foundation models on Amazon Bedrock and Vertex AI, and use our AI Safety Platform to choose.
  • Managed and serverless first. Containers on Fargate and Cloud Run, functions for event workloads, managed databases. Less to patch, more to build.
  • Security as code. Infrastructure, policies and guardrails are versioned, reviewed and deployed through pipelines.
  • Portability. Containerized services and open standards (FHIR, OpenTelemetry, Terraform) keep us free to place each workload where it fits best.
PRIMARY REGION · AWS
ca-central-1 (Montréal)
Prod
DR REGION · AWS
ca-west-1 (Calgary)
Standby
PRIMARY REGION · GCP
northamerica-northeast1 (Montréal)
Prod
SECONDARY · GCP
northamerica-northeast2 (Toronto)
Standby
ENCRYPTION
Customer-managed keys, TLS 1.2+
Enforced
Target deployment topology.
Reference architecture

The platform, layer by layer

Equivalent managed services on each cloud, so workloads can be placed by data-residency, model availability and cost.

Experience
Amazon CloudFront, AWS Amplify Hosting
Cloud CDN, Firebase Hosting
Identity & access
Amazon Cognito, IAM Identity Center
Identity Platform, Cloud IAM
Application services
Amazon ECS on Fargate, AWS Lambda, API Gateway
Cloud Run, Cloud Functions, API Gateway
Foundation models
Amazon Bedrock (with Guardrails)
Vertex AI (Gemini models)
Custom ML & speech
Amazon SageMaker, Amazon Transcribe
Vertex AI Training & Pipelines, Speech-to-Text
Data
Amazon RDS for PostgreSQL, Amazon S3, DynamoDB
Cloud SQL, Cloud Storage, BigQuery
Health data
AWS HealthLake (FHIR)
Cloud Healthcare API (FHIR)
Security & keys
AWS KMS, Secrets Manager, GuardDuty, Security Hub
Cloud KMS, Secret Manager, Security Command Center
Observability & audit
Amazon CloudWatch, AWS CloudTrail, AWS Config
Cloud Logging & Monitoring, Cloud Audit Logs
Delivery
Amazon ECR, AWS CodePipeline, CloudFormation / CDK
Artifact Registry, Cloud Build, Terraform
Workloads

How each product and platform uses the cloud

Our work is compute- and model-intensive: speech processing, model training, evaluation campaigns, red-team generation and runtime guardrails all run on managed cloud services.

  • Dementia DetectionSecure capture of speech and memory tasks, transcription, feature extraction and model training on consented research data.
    AWS: Transcribe, SageMaker, S3, HealthLake · GCP: Speech-to-Text, Vertex AI, Healthcare API
  • Dementia CareVoice companion, routines engine, notifications and care-team portal.
    AWS: Bedrock, Lambda, SNS, RDS · GCP: Gemini on Vertex AI, Cloud Run, Cloud SQL
  • AI Safety PlatformLarge-scale evaluation and red-team runs, scoring pipelines and a runtime guardrail service that screens every live interaction.
    AWS: Bedrock Guardrails, SageMaker, ECS Fargate, Lambda · GCP: Vertex AI, Cloud Run, BigQuery
  • AI Governance PlatformModel registry, risk register, consent ledger, release-gate workflows and a tamper-evident audit store.
    AWS: DynamoDB, Step Functions, S3 Object Lock, CloudTrail Lake · GCP: Firestore, Workflows, Cloud Storage retention lock, BigQuery
Roadmap

Where we're headed

Now · Foundation

Landing zones on AWS and Google Cloud, identity, CI/CD, and the first releases of our AI Safety and AI Governance platforms.

Next · Design partners

Pilots of the Dementia Care companion and consented research data collection for Dementia Detection, fully under both guardians.

Then · Validation & growth

Clinical validation of Dementia Detection, preparation for Health Canada medical-device review, and expansion into health guidance and mental-health support.

Security & privacy controls

How we protect data

  • Separate accounts and projects per environment and per sensitive workload, under organization-level guardrails
  • Encryption in transit and at rest with customer-managed keys; secrets never in code
  • Single sign-on with MFA for all staff; least-privilege, time-bound access to production
  • Centralized, immutable audit logs and continuous posture monitoring
  • Personal health information handled under PIPEDA and applicable provincial law, with consent records
  • Customer and participant data is never used to train general-purpose models without explicit consent