What Bob Is

Bob is a Slack-first assistant backed by Vertex AI Agent Engine. He orchestrates a team of specialist agents (the iam-* department) that handle ADK/Vertex-specific work:

  • ๐Ÿ” iam-adk - ADK pattern analysis and compliance
  • ๐Ÿšจ iam-issue - GitHub issue creation from findings
  • ๐Ÿ› ๏ธ iam-fix-plan - Fix planning and strategy
  • โš™๏ธ iam-fix-impl - Fix implementation
  • โœ… iam-qa - Quality assurance validation
  • ๐Ÿ“ iam-doc - Documentation generation
  • ๐Ÿงน iam-cleanup - Code cleanup and optimization
  • ๐Ÿ“š iam-index - Knowledge indexing and RAG

Designed as a reusable pattern: The architecture, documentation, and CI/CD pipeline are built to be copied into other organizations and repositories.

How Bob Works

๐Ÿ’ฌ
Slack User sends command
โ†’
๐ŸŒ
Cloud Run Gateway REST proxy
โ†’
๐Ÿค–
Vertex AI Agent Engine Bob orchestrator

Agent Coordination

  • Bob delegates complex tasks to iam-senior-adk-devops-lead (foreman agent)
  • Foreman coordinates 8 specialist agents for end-to-end workflows
  • Dual memory model: Vertex AI Session Service + Memory Bank (ready when enabled)
  • ARV (Agent Readiness Verification) gates ensure quality and safety
  • Portfolio-wide audits across multiple repositories

Architecture Diagrams

Repository Structure (v0.9.0 Agent Factory)

Canonical directory structure following ADK/Vertex Agent Engine best practices:

graph TB
    subgraph "bobs-brain/ (v0.9.0 Agent Factory)"
        A[agents/] --> A1[bob/]
        A --> A2[iam-senior-adk-devops-lead/]
        A --> A3[iam-* specialists/]

        B[service/] --> B1[a2a_gateway/]
        B --> B2[slack_webhook/]

        C[infra/terraform/] --> C1[agent_engine.tf]
        C --> C2[storage.tf]
        C --> C3[iam.tf]

        D[000-docs/] --> D1[Plans & AARs]
        D --> D2[Runbooks]
        D --> D3[Architecture docs]

        E[scripts/ci/] --> E1[check_nodrift.sh]
        E --> E2[arv checks]

        F[tests/] --> F1[Unit tests]
        F --> F2[Integration tests]
    end

    style A fill:#808080,color:#fff,stroke:#4d4d4d
    style B fill:#cccccc,color:#1a1a1a,stroke:#808080
    style C fill:#cccccc,color:#1a1a1a,stroke:#808080
    style D fill:#cccccc,color:#1a1a1a,stroke:#808080
    style E fill:#cccccc,color:#1a1a1a,stroke:#808080
    style F fill:#cccccc,color:#1a1a1a,stroke:#808080
                    

System Flow (Slack โ†’ Agent Engine)

How user requests flow from Slack to Bob and through the iam-* department:

sequenceDiagram
    participant U as User (Slack)
    participant S as Cloud Run Gateway
    participant B as Bob (Orchestrator)
    participant F as Foreman Agent
    participant I as iam-* Specialists
    participant V as Vertex AI / GCS

    U->>S: @Bob message
    S->>S: Validate webhook & auth
    S->>B: POST to Agent Engine
    B->>B: Dual memory: Session + Memory Bank
    B->>F: Delegate to iam-senior-adk-devops-lead
    F->>I: Coordinate iam-adk, iam-fix, iam-qa
    I->>V: Query Vertex AI Search / write to GCS
    V-->>I: Context & tools response
    I-->>F: Task results
    F-->>B: Consolidated report
    B->>S: Format response
    S->>U: Reply in Slack thread

    Note over B,F: A2A protocol for agent coordination
    Note over I,V: RAG with Vertex AI Search + ARV gates
                    

Deployment Flow (CI/CD with Hard Mode)

From code push to production via GitHub Actions and Agent Engine:

graph LR
    A[Git Push to main] --> B[GitHub Actions]
    B --> C{Drift Check R8}
    C -->|Violations| D[โŒ Block CI]
    C -->|Pass| E[Run ARV Checks]
    E --> F{ARV Pass?}
    F -->|No| G[โŒ Report Failures]
    F -->|Yes| H[ADK CLI Deploy]
    H --> I[Vertex AI Agent Engine]
    I --> J{Health Check}
    J -->|Success| K[โœ… Production Live]
    J -->|Fail| L[๐Ÿ”„ Rollback]

    style A fill:#808080,color:#fff,stroke:#4d4d4d
    style K fill:#4d4d4d,color:#fff,stroke:#1a1a1a
    style D fill:#999999,color:#fff,stroke:#4d4d4d
    style G fill:#999999,color:#fff,stroke:#4d4d4d
    style L fill:#b3b3b3,color:#1a1a1a,stroke:#808080
                    

Agent Department Model

How Bob delegates work to the foreman and specialist agents:

graph TD
    A[Bob - Global Orchestrator] --> B[iam-senior-adk-devops-lead]
    B --> C[iam-adk]
    B --> D[iam-issue]
    B --> E[iam-fix-plan]
    B --> F[iam-fix-impl]
    B --> G[iam-qa]
    B --> H[iam-doc]
    B --> I[iam-cleanup]
    B --> J[iam-index]

    C --> C1[ADK pattern analysis]
    D --> D1[Create GitHub issues]
    E --> E1[Design fix strategies]
    F --> F1[Implement fixes]
    G --> G1[Quality validation]
    H --> H1[Generate docs & AARs]
    I --> I1[Code optimization]
    J --> J1[Knowledge indexing]

    style A fill:#4d4d4d,color:#fff,stroke:#1a1a1a
    style B fill:#808080,color:#fff,stroke:#4d4d4d
    style C fill:#cccccc,color:#1a1a1a,stroke:#808080
    style D fill:#cccccc,color:#1a1a1a,stroke:#808080
    style E fill:#cccccc,color:#1a1a1a,stroke:#808080
    style F fill:#cccccc,color:#1a1a1a,stroke:#808080
    style G fill:#cccccc,color:#1a1a1a,stroke:#808080
    style H fill:#cccccc,color:#1a1a1a,stroke:#808080
    style I fill:#cccccc,color:#1a1a1a,stroke:#808080
    style J fill:#cccccc,color:#1a1a1a,stroke:#808080
                    

Tech Stack

Infrastructure

  • Google Cloud Run
  • Vertex AI Agent Engine
  • Cloud Storage (GCS)
  • Cloud Logging & Trace

AI & Knowledge

  • Google ADK (Agent Development Kit)
  • Vertex AI Search (RAG ready)
  • Gemini 2.0 Flash
  • Org-wide knowledge hub

DevOps & Quality

  • GitHub Actions CI/CD
  • ARV readiness verification
  • Terraform IaC
  • Workload Identity Federation

For Teams & Organizations

This repository is my reference implementation for building real multi-agent systems on Vertex AI using Google's ADK. It demonstrates production patterns for:

  • ๐Ÿข Slack copilots integrated with your organization's workflows
  • ๐Ÿ”ง ADK-compliant agent departments with specialized roles
  • ๐Ÿ“Š Portfolio-wide automation across multiple GitHub repositories
  • ๐Ÿ” Enterprise-grade security with WIF and proper IAM
  • ๐Ÿ“ˆ Scalable architecture ready for production workloads

Need something similar for your organization?
Intent Solutions can help design and implement custom multi-agent systems, Slack assistants, and AI-powered automation tailored to your needs.

Learn more at intentsolutions.io โ†’

Documentation & Resources