Google ADK Agentic Solution: How I Published a Blog Post Using Agents

A Google ADK multi-agent app used to research

Author

Dr. Aruna Kishore Veleti
August 15, 2026

This post describes a Google ADK multi-agent app used to research Google Cloud documentation and draft a technical article. The app does not implement offer qualification or order creation.

The published article below is the output of the above Github codebase (one ADK web session, then copy to WordPress):

What the codebase does

dev-signal is a Python Google ADK app. A root orchestrator (root_orchestrator) delegates to three specialists:

Agent

Role

reddit_scanner

Recent high-engagement questions (Reddit MCP; needs Reddit Data API access)

gcp_expert

Official GCP docs via Developer Knowledge MCP, plus web search and Reddit

blog_drafter

Draft from stored research; optional header image (Nano Banana / Gemini image → Cloud Storage)

gcp_expert writes findings to session state as technical_research_findings. blog_drafter reads that state, re-checks facts with Developer Knowledge, then drafts.

Long-term preferences use Vertex AI Memory Bank when you run test_local.py. The ADK web UI uses in-memory sessions only.

Agentic Architecture Applied - Runtime representation while executing (how it shows in your laptop when you this codebase http://127.0.0.1:8000/dev-ui/?app=dev_signal_agent&userId=user&session=75a7a218-610b-4ecd-a822-9b5649c66aeb )

How the draft was produced

  1. Configure .env (GCP project, GCS bucket, Developer Knowledge API key).

  2. Authenticate with gcloud (user login + application-default credentials).

  3. Run uv run adk web, open http://127.0.0.1:8000, select dev_signal_agent.

  4. Prompt through research → draft → header image → revisions.

  5. Save the draft in the repo (Docs/scaling-intelligent-commerce-adk-cloud-run.md) and publish it on Product Cognizant.

Prompts used (in order)

  1. Hi

  2. Qualify offers and create orders

  3. How do I deploy an ADK agent to Cloud Run?

  4. Draft a blog post based on this Cloud Run ADK research

  5. Yes (header image), then generate the header image now

  6. Now give me the full blog post

  7. Production should use enterprise CI/CD (GitHub Actions / Artifact Registry), not adk deploy from a laptop

  8. Frame it as git feature → develop → release, then Dev / QA / Staging / Integration / Pre-prod / Prod / Prod-passive failover

  9. You drifted off the original topic — put it back

  10. Keep the title Scaling Intelligent eCommerce: Deploying ADK Agents to Google Cloud Run; no marketing copy; steps and focus areas only

  11. Don’t rewrite the earlier full draft; improve it

  12. After the opening line, add Business Domain, Business Processes, Why Agentic, Solution Approach, then keep the rest

Prompt 2 set the topic of the output post (offers and orders). It is not a feature of this repository.

What the output post covers

That article is a Cloud Run ADK deploy guide for a multi-agent layout: Qualification Agent (reasoning) and Order Agent (ERP/CRM tools). It includes local root_agent structure, build-once / deploy-many CI/CD, --session_service_uri, IAM (roles/aiplatform.user, roles/secretmanager.secretAccessor), and active/passive failover.