v1.0 | Last updated: 2026-05-26
The Cytognosis infrastructure follows extreme simplicity, serverless-first, and HIPAA-ready architectural choices. Three GCP projects, three domain TLDs, and a composable container framework serve all platform needs.
``mermaid`
flowchart TB
subgraph GCP["Google Cloud Platform"]
infra["cytognosis-infrastructure
DNS, IAM, legacy buckets"]
phi["cytognosis-phi-prod
Website, services, PHI"]
data["cytognosis-data
Data platform, analytics"]
end
subgraph Storage["Cloud Storage"]
datahub["gs://cytognosis-data-hub/
DVC cache, datasets, shared"]
phiprod["gs://cytognosis-phi-prod/
HIPAA-compliant PHI data"]
end
subgraph Compute["Compute"]
cloudrun["Cloud Run
Website, APIs"]
cytohost["cytohost VM
Core services"]
end
infra --> datahub
phi --> phiprod
phi --> cloudrun
infra --> cytohost
| Project ID | Purpose | Classification |
|------------|---------|----------------|
| cytognosis-infrastructure | DNS zones, IAM root, legacy buckets | Management |cytognosis-phi-prod
| | Website, services, user data | Sensitive (PHI) |cytognosis-data
| | Data platform, analytics | Regulated |
Three primary apex domains:
| Domain | Purpose | Registrar |
|--------|---------|-----------|
| cytognosis.org | Primary canonical | Squarespace |cytognosis.com
| | Secondary canonical | Squarespace |cytognosis.ai
| | Technology canonical | Squarespace |
DNS routing uses a dual-zone topology in Cloud DNS for backward compatibility:
, cg-com, cg-ai, com-zone, ai-zonemermaid
flowchart LR
subgraph Internet["Public Internet"]
users["Users"]
end
subgraph CloudRun["Cloud Run (Serverless)"]
website["cytognosis.org
Main website"]
end
subgraph Cytohost["cytohost VM"]
caddy["Caddy
Reverse proxy"]
core["Core Stack
Cal.com, Excalidraw,
Mermaid, Logseq, MLflow"]
ondemand["On-demand
Neo4j, SurrealDB,
Jupyter"]
end
users --> website
users --> caddy --> core
caddy --> ondemand
`
On-Demand Service Model
Services default OFF and start when needed. The
core stack runs always-on behind Caddy. Everything else (Neo4j, Jupyter, SurrealDB) starts on demand via stack_manager.py.
| Stack | Services | Mode |
|-------|----------|------|
| core | Caddy, Cal.com, Excalidraw, Mermaid, Logseq, MLflow | Always-on |
| research | Neo4j, Jupyter, MLflow | On-demand |
| neo4j-only | Neo4j | On-demand |
See Container Framework for full details.
Data Layer
`
gs://cytognosis-data-hub/
├── dvc-cache/ Content-addressed DVC cache (shared)
├── processed/cytos/dvc/ Cytos project DVC remote
├── purdue/ Collaborator workspace
├── shared/
│ ├── soma/ TileDB-SOMA exports
│ ├── gwas/ GWAS summary statistics
│ ├── embeddings/ Feature vectors
│ └── reference/ GRCh38, GENCODE GTFs
├── public-mirror/ Public dataset mirrors
└── manifests/ Dataset manifest JSONs
`
See GCP Setup and DVC Strategy for details.
Provenance Stack
`
L0: DVC + VFS Content-addressed hashes, SWHID for code
L1: redun / Nextflow Workflow DAG lineage
L2: Artifact Registry Queryable metadata
L3: MLflow Experiment tracking
L4: RO-Crate FAIR publication packages
``