DAT409 - Hybrid Search with Aurora PostgreSQL for MCP Retrieval

⚠️ Important Notice: For demonstration and educational purposes only. Not intended for production use.

🚀 Quick Start

Workshop Duration: 60 minutes | Lab 1: 25 min | Lab 2: 20 min

Build hybrid search with Aurora PostgreSQL, pgvector, and MCP for natural language database queries.

📁 Repository Structure

├── lab1-hybrid-search/
│   ├── notebook/
│   │   └── dat409-hybrid-search-notebook.ipynb  # Main lab notebook
│   ├── data/
│   │   └── amazon-products.csv                  # 21,704 products
│   └── requirements.txt                         # Lab 1 dependencies
├── lab2-mcp-agent/
│   ├── streamlit_app.py                         # Streamlit demo app
│   ├── test_personas.sh                         # RLS testing script
│   ├── mcp_config.json                          # MCP configuration
│   └── requirements.txt                         # Lab 2 dependencies
├── scripts/
│   ├── bootstrap-code-editor.sh                 # Infrastructure setup
│   ├── setup-database.sh                        # Database & data loading
│   └── setup/                                   # Helper scripts
├── cfn/                                         # CloudFormation templates
├── env_example                                  # Environment template
└── README.md                                    # This file

🎯 Labs

Lab 1: Hybrid Search (25 min)

Combine vector similarity (pgvector + HNSW), full-text search, and fuzzy matching (pg_trgm) with 21,704 products.

cd /workshop/lab1-hybrid-search/notebook
# Open dat409-hybrid-search-notebook.ipynb

Lab 2: MCP & RLS (20 min)

Natural language queries with Model Context Protocol and Row-Level Security for multi-tenant access.

cd /workshop/lab2-mcp-agent
./test_personas.sh  # Test customer, support, product manager personas
streamlit run streamlit_app.py  # Optional demo

🛠️ Prerequisites

• AWS Account with Aurora PostgreSQL 17.5 + Amazon Bedrock (Cohere Embed English v3)
• Python 3.13 + Jupyter Notebook

🚦 Getting Started

Participants: Access Code Editor via CloudFront URL → Open Lab 1 notebook in /workshop/lab1-hybrid-search/notebook/

Instructors: See DEPLOYMENT_SEQUENCE.md for setup.

🔧 Technical Stack

• Database: Aurora PostgreSQL 17.5 with pgvector extension
• Embeddings: Cohere Embed English v3 (1024 dimensions)
• Search: HNSW vector index + GIN full-text + pg_trgm fuzzy
• MCP: Model Context Protocol for database access
• AI Agent: Strands Agent with Claude Sonnet 4 + MCP tools
• RLS: Row-Level Security for persona-based access
• Python: 3.13 with pandas, psycopg, boto3, streamlit

🤖 MCP Agent Architecture

The workshop demonstrates natural language database queries using a Strands Agent with MCP tools:

User Query → Strands Agent (Claude Sonnet 4) → MCP Client → Aurora PostgreSQL (Data API)

Key Components:

• Strands Agent: AI agent framework with tool-calling capabilities
• MCP Client: Provides standardized database access tools via awslabs.postgres-mcp-server
• Claude Sonnet 4: Interprets queries and decides which MCP tools to call
• Aurora Data API: Serverless database access using cluster ARN + secret ARN

Why This Pattern?

• Agent uses admin access via Data API for intelligent cross-schema queries
• Application-level authorization handles security (typical production pattern for AI agents)
• MCP provides standardized, reusable database tools
• Enables natural language → SQL translation with context awareness

📚 Resources

• DEPLOYMENT_SEQUENCE.md | Aurora PostgreSQL | pgvector | MCP

⭐ Like This Workshop?

If you find this workshop helpful, please consider:

• ⭐ Star this repository to show your support!
• 🍴 Fork it to customize for your own use cases
• 🐛 Report issues to help us improve
• 💡 Submit pull requests with enhancements or fixes
• 📢 Share it with your colleagues and community

Your feedback and contributions help make this workshop better for everyone!

🤝 Contributing

Contributions welcome! Documentation, bug fixes, features, tests, or feedback. See CONTRIBUTING.md.

📄 License & Security

MIT-0 License. See LICENSE | Security: CONTRIBUTING.md

---

© 2025 Shayon Sanyal, Principal Solutions Architect, AWS