How googlecloudplatform/genai-for-marketing works
Project type and purpose
• This is a full-stack, cloud-native marketing solution that uses Google Cloud’s Generative AI services.
• The Angular-based frontend drives various marketing use cases (insights, content generation, trendspotting, search).
• The Python backend (in backend_apis/) exposes REST endpoints, orchestrates calls to BigQuery, Vertex AI (text+vision models), Firebase, GCS, Workspace APIs and Looker.
• Terraform (infra/) codifies all cloud resources.
File structure
• Top-level directories:
• backend_apis/ – Python app that wraps AI, database and external service calls
• frontend/ – Angular application (components, services, routing)
• infra/ – Terraform modules, BigQuery schemas, IAM, deployment scripts
• notebooks/ – Jupyter demos illustrating the AI workflows
• templates/ – slide/doc templates for Workspace integration
• Key config files:
• app.yaml – App Engine config for frontend (or gRPC proxy)
• backend_apis/Dockerfile & requirements.txt – containerizing the Python APIs
• infra/*.tf – Terraform resources: Cloud Run/App Engine, BigQuery, IAM, GCS buckets, Vertex AI endpoints
• infra/templates/config.toml.tftpl → config.toml for AI prompts/models
README for additional insights
• Frontend UI
• Backend APIs
• BigQuery
• Vertex AI (Search + Generative Models)
• Google Workspace and Looker
• It calls out:
• Marketing Insights (Looker dashboards)
• Audience Q&A (natural language→SQL)
• Trendspotting (Vertex+BigQuery+news summarization)
• Content Search (Vertex AI Search)
• Content Generation (Vertex Text & Vision models)
• Workspace integration (Docs, Slides, Sheets)
Essential architectural elements – don’t miss
• Stateless containerized API calling multiple AI endpoints
• Angular routing to various “demos” (trendspotting, copy generation)
• BigQuery as the analytics store feeding dashboards and Q&A
• Vertex AI Search layer in front of content index
• Google Workspace integration for automated asset delivery
• Terraform provisioning, sample tfvars by vertical