koborin.ai

Technical garden for exploring AI, cloud architecture, and continuous learning.

Astro ( ) runs on Cloud Run behind a global HTTPS load balancer, and the entire stack (app + infra) lives in this monorepo with Pulumi (Go).

Architecture

Dev/Prod share the same HTTPS load balancer and Artifact Registry; only Cloud Run scaling/access policies differ.

---
title: "Google Cloud Project"
---
flowchart LR
    subgraph PULUMI_STATE["Pulumi Backend State - GCS"]
        STATE_SHARED["shared"]
        STATE_DEV["dev"]
        STATE_PROD["prod"]
    end
    
    subgraph SHARED["Shared Resources"]
        APIS["APIs"]
        ARTIFACT["Artifact Registry"]
        WIF["Workload Identity<br/>Pool/Provider"]
    end
    
    subgraph DNS["Cloudflare DNS (manual)"]
        DEV_DOMAIN["dev.koborin.ai<br/>A record"]
        PROD_DOMAIN["koborin.ai<br/>A record"]
    end
    
    subgraph LB["Unified HTTPS LB - shared"]
        STATIC_IP["Static IP<br/>PREMIUM tier"]
        SSL_CERT["Managed SSL Cert<br/>multi-domain"]
        URL_MAP["URL Map<br/>host routing"]
        HTTPS_PROXY["HTTPS Proxy"]
    end
    
    subgraph DEV_BACKEND["Dev Backend"]
        DEV_IAP["IAP<br/>+ X-Robots-Tag"]
        DEV_NEG["Serverless NEG"]
        DEV_CR["Cloud Run<br/>koborin-ai-web-dev (Astro)<br/>(LB-only ingress)"]
    end
    
    subgraph PROD_BACKEND["Prod Backend"]
        PROD_NEG["Serverless NEG"]
        PROD_CR["Cloud Run<br/>koborin-ai-web-prod (Astro)<br/>(LB-only ingress)"]
    end
    
    subgraph CICD["GitHub Actions"]
        GH_WIF["Workload Identity<br/>Federation"]
        PULUMI_SA["Pulumi SA"]
    end
    
    STATE_SHARED -.-> SHARED
    STATE_SHARED -.-> LB
    STATE_DEV -.-> DEV_BACKEND
    STATE_PROD -.-> PROD_BACKEND
    
    ARTIFACT -.->|"Container Image"| DEV_CR
    ARTIFACT -.->|"Container Image"| PROD_CR
    
    DEV_DOMAIN --> STATIC_IP
    PROD_DOMAIN --> STATIC_IP
    STATIC_IP --> SSL_CERT
    SSL_CERT --> HTTPS_PROXY
    HTTPS_PROXY --> URL_MAP
    
    URL_MAP -->|"dev host"| DEV_IAP
    URL_MAP -->|"prod host"| PROD_NEG
    
    DEV_IAP --> DEV_NEG
    DEV_NEG --> DEV_CR
    PROD_NEG --> PROD_CR
    
    GH_WIF -.->|"Authenticate"| WIF
    WIF -.->|"Impersonate"| PULUMI_SA
    PULUMI_SA -.->|"Preview/Up"| PULUMI_STATE
    
    style PULUMI_STATE fill:#E8E8E8,color:#000,stroke:#666,stroke-width:2px
    style STATE_SHARED fill:#F5F5F5,color:#000
    style STATE_DEV fill:#F5F5F5,color:#000
    style STATE_PROD fill:#F5F5F5,color:#000
    style APIS fill:#9E9E9E,color:#fff
    style ARTIFACT fill:#9E9E9E,color:#fff
    style WIF fill:#9E9E9E,color:#fff
    style STATIC_IP fill:#00B6AC,color:#fff
    style SSL_CERT fill:#00B6AC,color:#fff
    style URL_MAP fill:#00B6AC,color:#fff
    style HTTPS_PROXY fill:#00B6AC,color:#fff
    style DEV_IAP fill:#F76560,color:#fff
    style DEV_CR fill:#4A90E2,color:#fff
    style PROD_CR fill:#4A90E2,color:#fff
    style GH_WIF fill:#FFB74D,color:#000
    style PULUMI_SA fill:#FFB74D,color:#000

DNS is hosted in Cloudflare. Pulumi does not manage DNS records; add/update koborin.ai / dev.koborin.ai A records manually whenever the load balancer IP changes.

Environment matrix

CI/CD

Infrastructure and application deploys are each handled by dedicated GitHub Actions workflows using Workload Identity Federation.

flowchart LR
    subgraph GitHub Actions
        planInfra[plan-infra.yml]
        releaseInfra[release-infra.yml]
        appCI[app-ci.yml]
        appRelease[app-release.yml]
    end

    subgraph Cloud Build
        buildApp[Docker Build\n+ Artifact Registry]
    end

    subgraph Pulumi
        sharedStack[Shared Stack]
        envStacks[Dev/Prod Stacks]
    end

    planInfra --> sharedStack
    releaseInfra --> sharedStack
    releaseInfra --> envStacks
    appCI -->|PR validation| GitHub
    appRelease --> buildApp --> envStacks

Tech Stack

• Frontend: Astro with Starlight (documentation theme), TypeScript, Tailwind CSS.
• Content Management: MDX stored under app/src/content/docs/ within git. Frontmatter is validated via Zod schemas (from Starlight) to keep metadata type-safe. Drafts can be marked with draft: true in frontmatter.
• Analytics & o11y:
  • Google Analytics 4 for baseline PV/engagement.
  • Optional custom /api/track endpoint writing to Cloud Logging → BigQuery for privacy-friendly metrics.
  • Cloud Monitoring dashboards + alert policies (via Pulumi) for Cloud Run metrics.
• Infrastructure: Pulumi (Go) targeting Google Cloud.
• CI/CD: GitHub Actions with Workload Identity. plan-infra.yml / release-infra.yml drive infra, app-ci.yml / app-release.yml handle the Astro app.
• Testing: Vitest for app tests, TypeScript compilation for infra, Playwright for future E2E if needed.
• LLM Context: Machine-readable llms.txt files for AI assistants. Auto-generated at build time.

LLM Context Files (llms.txt)

The site provides structured context files for LLMs at https://koborin.ai/llms.txt.

These files are auto-generated at build time from Content Collections. Articles with draft: true are excluded. No runtime overhead.

Repository Layout (planned)

.
├── .gcloudignore                  # Excludes files from Cloud Build upload
├── app/                           # Astro + Starlight application (Static, nginx)
│   ├── cloudbuild.yaml            # Cloud Build config for Docker build
│   ├── src/
│   │   ├── assets/               # Images organized by category
│   │   │   ├── _shared/          # Common assets (header logo)
│   │   │   ├── tech/             # Tech article images
│   │   │   ├── life/             # Life article images
│   │   │   └── about-me/         # About me article images
│   │   ├── content/
│   │   │   ├── docs/             # MDX documentation pages (Starlight)
│   │   │   └── config.ts         # Content Collections schema
│   │   ├── utils/
│   │   │   └── llms.ts           # Shared logic for llms.txt generation
│   │   ├── pages/                # Astro endpoints (llms.txt, RSS feeds)
│   │   └── styles/
│   │       └── custom.css        # Custom CSS overrides (logo sizing, etc.)
│   ├── public/
│   │   ├── favicon.png           # Browser tab icon
│   │   └── robots.txt
│   ├── nginx/
│   │   └── nginx.conf            # nginx configuration for static serving
│   ├── Dockerfile                # Multi-stage build (node → nginx:alpine)
│   └── astro.config.mjs          # Starlight integration config
├── docs/                          # Architecture notes, contact-flow specs, etc.
├── infra/                         # Pulumi Go stacks (shared/dev/prod)
│   ├── main.go                   # Entry point
│   ├── config.go                 # Configuration helpers
│   ├── stacks/                   # Stack definitions
│   │   ├── shared.go             # Shared resources (LB, APIs, WIF)
│   │   ├── dev.go                # Dev Cloud Run
│   │   └── prod.go               # Prod Cloud Run
│   ├── Pulumi.yaml               # Project configuration
│   ├── go.mod                    # Go module dependencies
│   └── go.sum                    # Go dependency checksums
├── .claude-plugin/                # Plugin marketplace manifest
│   └── marketplace.json           # Registry of published plugins
├── plugins/                       # Published Claude Code plugins
│   ├── agent-team-fullstack/      # Agent Team orchestration plugin
│   └── mermaid-diagram/           # Mermaid diagram workflow plugin
├── .github/workflows/             # CI pipelines (plan/apply, app deploy, plugin validation)
├── README.md                      # This file
└── AGENTS.md                      # English operations guide for collaborators

Brand Assets

Logo sizing is customized via app/src/styles/custom.css (.site-title img selector).

Image Optimization (Automatic)

For performance, images are automatically converted to WebP format. Authors can use PNG/JPEG normally - optimization happens during build/deploy.

Example workflow for OG images:

1. Place image: app/public/og/my-article.png
2. Frontmatter: ogImage: /og/my-article.png
3. Article display: ![](/og/my-article.png)

That's it! The CI pipeline (app/scripts/optimize-og-images.sh) generates WebP versions, and nginx automatically serves them.

Workflow Overview

1. Infra changes: edit Pulumi Go stacks → go build ./... && go vet ./... → open PR → GitHub Actions runs preview → reviewer approves → merge triggers apply on the right environment.
2. App changes: edit Astro/MDX → npm run lint && npm run test && npm run typecheck && npm run check-images && npm run build → PR triggers app-ci.yml → merge to main (or tag app-v*) triggers app-release.yml which builds the container, pushes to Artifact Registry, and feeds the new image to Pulumi.
3. Content-only updates: modify MDX under app/src/content/docs/, update frontmatter (title, description), run npm run lint, open PR. Mark drafts with draft: true in frontmatter to exclude from production builds.

Adding New Content

To add a new article or page:

1. Create MDX file under app/src/content/docs/ (or subdirectory for categories):

   # Single page
   app/src/content/docs/my-article.mdx
   
   # Categorized page
   app/src/content/docs/blog/my-post.mdx
   

2. Add frontmatter with required fields:

   ---
   title: My Article Title
   description: Brief description of the article
   publishedAt: 2025-01-02
   ---
   

   • publishedAt: Set the publish date manually (YYYY-MM-DD format).
   • Updated date is automatically extracted from Git history at build time.

3. Update sidebar in app/src/sidebar.ts:

   export const sidebar = [
     // ... existing items
     {
       label: "Blog",
       items: [
         { label: "My Post", slug: "blog/my-post" },
       ],
     },
   ];
   

4. Build and test locally before pushing.

Release Strategy

• Infra applies use infra-v* tags to trigger release-infra.yml. Tag the repo after merging infra PRs even if app work is still ongoing; this ensures the latest load balancer/stateful resources are deployed before app images roll out.
• App deploys use app-v* tags to drive app-release.yml. Tagging after a successful main merge guarantees that the latest container image is built and the Cloud Run service is updated via Pulumi.
• GitHub release notes are generated via .github/release.yml. Label each PR with app, infra, pulumi, feature, bug, or doc so the notes stay segmented by domain; apply the ignore label to omit a PR entirely.

Local Setup (once the app repo is initialized)

# Node.js >= 20, npm 10 recommended
npm install

# Run Astro dev server (app directory)
npm run dev --prefix app

# Build infrastructure (infra directory)
cd infra && go build ./...

Infrastructure Dev Notes

• Pulumi stacks are located in infra/stacks/.
• GCP provider version is managed via github.com/pulumi/pulumi-gcp/sdk Go module.
• Each stack uses a GCS backend with automatic state management per stack name.

Shared Stack

• API enablement: Run, Compute, IAM, Artifact Registry, IAP, Monitoring, Logging, Certificate Manager.
• Artifact Registry: Container images repository (koborin-ai-web).
• Global static IP: PREMIUM tier for HTTPS load balancer.
• Managed SSL Certificate: Multi-domain (koborin.ai, dev.koborin.ai).
• HTTPS Load Balancer:
  • Serverless NEGs (dev/prod) referencing Cloud Run services by name.
  • Backend Services:
    • Dev: IAP enabled + X-Robots-Tag: noindex, nofollow header.
    • Prod: No IAP, logging enabled.
  • URL Map (host-based routing).
  • Target HTTPS Proxy.
  • Global Forwarding Rule.
• Workload Identity Federation:
  • Pool: github-actions-pool.
  • Provider: actions-firebase-provider (OIDC issuer: https://token.actions.githubusercontent.com).
  • Service Account: github-actions-service@{project}.iam.gserviceaccount.com.
  • IAM binding: Subject-based binding for repository nozomi-koborinai/koborin-ai.
  • Project IAM roles (Artifact Registry, Run, Compute, IAM, etc.) granted to the Pulumi SA.
• DNS: Records live in Cloudflare and are managed manually (A records point to the LB IP).

Dev Stack

• Cloud Run Service: koborin-ai-web-dev.
  • Ingress: INGRESS_TRAFFIC_INTERNAL_LOAD_BALANCER (LB-only access).
  • Execution Environment: Gen2.
  • Environment Variables:
    • NODE_ENV=development (runtime mode).
    • NEXT_PUBLIC_ENV=dev (client-side environment identifier).
  • Scaling: Min 0, Max 1.

Prod Stack

• Cloud Run Service: koborin-ai-web-prod.
  • Ingress: INGRESS_TRAFFIC_INTERNAL_LOAD_BALANCER (LB-only access).
  • Execution Environment: Gen2.
  • Environment Variables:
    • NODE_ENV=production (runtime mode).
    • NEXT_PUBLIC_ENV=prod (client-side environment identifier).
  • Scaling: Min 0, Max 10.

Contact & Analytics Design (planned)

• Contact form will post to /api/contact (Astro API Route) with:
  • Payload validation (Zod), reCAPTCHA enforcement, structured logging to Cloud Logging.
  • Notification via SendGrid or Gmail API (configured via Secret Manager).
• /api/track endpoint will receive custom events and forward to Cloud Logging/BigQuery.
• GA4 integration via gtag.js (prod only, injected at build time via PUBLIC_GA_MEASUREMENT_ID).

Environment Variables

The workflow creates a .env file with PUBLIC_GA_MEASUREMENT_ID only for prod builds. Dev builds do not include GA4 tracking.

Documentation

• README.md: quickstart + architectural highlights (this file).
• AGENTS.md: contribution workflow, review checklist, release rules, IaC philosophy.