A complete end-to-end Agentic Data Application built on Google Cloud Platform — transforming unstructured dark data into a production-grade AI agent system with real-time HTAP capabilities, LLM evaluation, and conversational analytics.
FroyoOS is a multi-codelab certification project demonstrating advanced Google Cloud skills across 5 interconnected codelabs. The system processes raw PDF documents through an AI pipeline, federates data across AlloyDB and BigQuery, deploys a multi-agent application, evaluates AI agents using LLM-as-a-Judge, and exposes conversational analytics through Antigravity IDE.
┌──────────────────────────────────────────────────────────────────┐
│ FroyoOS Architecture │
├──────────────────────────────────────────────────────────────────┤
│ │
│ PDFs ──► Dataplex DataScan ──► BigQuery (froyo_data) │
│ │ │
│ ▼ │
│ AlloyDB (bigquery_fdw) │
│ │ │
│ ▼ │
│ MCP Toolbox (Cloud Run) │
│ │ │
│ ▼ │
│ FroyoOS Agent App (Gemini) │
│ ├── check_allergens tool │
│ └── place_order tool │
│ │ │
│ ▼ │
│ LLM-as-a-Judge Evaluation Framework │
│ │ │
│ ▼ │
│ Antigravity IDE + BigQuery Conversational Agent │
└──────────────────────────────────────────────────────────────────┘
| # | Codelab | Key Technologies | Status |
|---|---|---|---|
| 1 | Dark Data to Structured Gold | Dataplex, BigQuery, GCS, Gemini | ✅ Complete |
| 2 | AlloyDB-BigQuery Zero ETL Federation | AlloyDB, bigquery_fdw, VPC | ✅ Complete |
| 3 | Agentic App with AlloyDB HTAP | MCP Toolbox, Cloud Run, Gemini ADK | ✅ Complete |
| 4 | Agent Evaluation with Gemini | LLM-as-a-Judge, Vertex AI, ADK Eval | ✅ Complete |
| 5 | Data Agents in Antigravity IDE & BigQuery | Antigravity IDE, BQ Conversational Agent | ✅ Complete |
froyo-agentic-cloud/
├── README.md # This file
├── CodeLab_1/
│ ├── README.md # Codelab 1 documentation
│ └── setup.sh # Setup script
├── CodeLab_2/
│ ├── README.md # Codelab 2 documentation
│ ├── setup_alloydb.sh # AlloyDB provisioning script
│ └── federation.sql # BigQuery federation SQL
├── CodeLab_3/
│ ├── README.md # Codelab 3 documentation
│ ├── tools.yaml # MCP Toolbox configuration
│ ├── deploy.sh # Cloud Run deployment script
│ └── app-nobill.py # Agent application
├── CodeLab_4/
│ ├── README.md # Codelab 4 documentation
│ ├── agent_eval.py # Evaluation script
│ └── eval_results.md # Sample evaluation results
├── CodeLab_5/
│ ├── README.md # Codelab 5 documentation
│ ├── GEMINI.md # Antigravity IDE context file
│ ├── load_tables.sh # BigQuery table loading script
│ └── bq_agent_config.json # BigQuery agent configuration
├── Screenshots/
└── .github/
└── workflows/
└── validate.yml # CI validation workflow
| Category | Technology |
|---|---|
| Database | AlloyDB (PostgreSQL 15), BigQuery |
| AI/ML | Gemini 2.0 Flash, Vertex AI, Google ADK |
| Infrastructure | Cloud Run, VPC, Private Service Connect |
| Data Pipeline | Dataplex DataScan, GCS, BigQuery Federation |
| Agent Framework | MCP Toolbox for Databases, LangChain |
| IDE | Antigravity IDE, Google Cloud Data Agent Kit |
| Security | Secret Manager, IAM, VPC Egress Controls |
| Language | Python 3.12, SQL, YAML |
- Google Cloud Project with billing enabled
gcloudCLI installed and configured- Python 3.12+
- AlloyDB cluster running
# Clone the repository
git clone https://github.com/YOUR_USERNAME/froyo-agentic-cloud.git
cd froyo-agentic-cloud
# Set your project
export PROJECT_ID="your-project-id"
gcloud config set project $PROJECT_ID
# Run Codelab 3 setup (requires AlloyDB + BigQuery from Codelabs 1 & 2)
cd codelab-3-agent-app
chmod +x deploy.sh
./deploy.sh- Zero-ETL Federation: Connected AlloyDB to BigQuery using
bigquery_fdwextension eliminating data duplication - HTAP Architecture: Hybrid Transactional/Analytical Processing — live orders in AlloyDB, analytics in BigQuery
- Production-grade Security: MCP Toolbox with parameterized SQL prevents SQL injection
- LLM Evaluation: Implemented LLM-as-a-Judge pattern scoring tool accuracy and groundedness
- Conversational Analytics: Natural language interface to complex multi-table BigQuery queries
| Metric | Score |
|---|---|
| Tool Routing Accuracy | 1.0 / 1.0 ✅ |
| Response Groundedness | 0.5 / 1.0 ✅ |
| SQL Injection Prevention | PASS ✅ |
| Jailbreak Resistance | PASS ✅ |
Dhyan Jain
- LinkedIn: linkedin.com/in/dhyanjain2701
- GitHub: github.com/DreamRealization
MIT License