HDB Multi-Agent Advisor

A phased multi-agent system for HDB property recommendations using AWS Bedrock AgentCore and the official AWS multi-agent pattern.

Phase 3: Cost-Optimized Multi-Agent System (Current)

Complete multi-agent system with intelligent routing for cost optimization between PostgreSQL and Athena.

Architecture

graph TD
    A[User] --> B[Web Interface]
    B --> C[API Gateway]
    C --> D[Chat Handler Lambda]
    D --> E[AgentCore Supervisor<br/>Amazon Nova Lite]
    
    E --> F{Query Routing}
    F -->|Demographics| G[Demographics Agent]
    F -->|Complex Analytics| H[Property Agent]
    F -->|Cost-Optimized| I[PostgreSQL Agent]
    
    G --> J[Data Commons API]
    J --> K[Singapore Gov Data]
    
    H --> L[Athena API Lambda]
    L --> M[Amazon Athena]
    M --> N[S3 HDB Data]
    
    I --> O[PostgreSQL API Lambda]
    O --> P[RDS Proxy]
    P --> Q[PostgreSQL RDS]
    Q --> R[HDB Resale Data<br/>2017+ Transactions]
    
    G --> S[Response]
    H --> S
    I --> S
    S --> D
    D --> C
    C --> B
    B --> A

Components:

• Supervisor Agent: Intelligent query routing using AWS AgentCore with Amazon Nova Lite
• Demographics Agent: Singapore population data via Data Commons API
• Property Agent: Complex HDB analytics via Athena queries (for statistical analysis)
• PostgreSQL Agent: Cost-optimized HDB queries via RDS Proxy (70% cheaper than Athena)
• Dual Data Sources: Same HDB data available via both PostgreSQL (fast) and Athena (complex analytics)

Capabilities & Examples

Project Structure

hdb-multiagent-advisor/
├── agent/
│   ├── supervisor.py             # ✅ Routes queries with Amazon Nova Lite
│   └── agents/
│       ├── demographics_agent.py # ✅ Singapore population & statistics
│       ├── property_agent.py     # ✅ Complex HDB analytics via Athena
│       └── postgres_agent.py     # ✅ Cost-optimized HDB queries
├── infrastructure/
│   ├── phase1-stack.yaml         # Web interface + chat handler
│   ├── phase2-stack.yaml         # Athena API + HDB database
│   ├── phase3-stack.yaml         # PostgreSQL API + RDS Proxy
│   └── cloudfront-stack.yaml     # CloudFront distribution (auto-generated)
├── lambda/
│   ├── chat_handler/             # Web interface backend
│   ├── athena_complex_query/     # HDB data API endpoints (complex)
│   └── postgres_query/           # HDB data API endpoints (cost-optimized)
├── web/index.html                # Chat interface with 3 phase examples
├── deploy-phase1.ps1             # Deploy web interface
├── deploy-phase2.ps1             # Deploy Athena database
├── deploy-phase3.ps1             # Deploy PostgreSQL integration
├── deploy-cloudfront.ps1         # Deploy CloudFront distribution
└── setup.ps1                     # Environment setup

Migration Path Completed

✅ Phase 1: Foundation (Completed)

• Single HDB assistant with Data Commons integration
• Basic chat interface and memory persistence

✅ Phase 2: Multi-Agent HDB System (Completed)

• Property Agent: HDB search, statistics, price trends using Athena + S3
• Demographics Agent: Singapore population and statistics via Data Commons
• Supervisor Agent: Intelligent query routing between specialized agents
• Athena API: Full REST API for HDB resale data (2017 onwards)
• Web Interface: Real-time chat with multi-agent responses

✅ Phase 3: Cost-Optimized Multi-Agent System (Current)

• PostgreSQL Agent: Cost-optimized HDB queries via RDS Proxy (70% cheaper)
• Intelligent Cost Routing: Automatic selection between PostgreSQL and Athena
• RDS Proxy Integration: Secure, scalable database connections
• Cost Optimization: PostgreSQL for frequent queries, Athena for complex analytics
• Dual Data Sources: Same HDB data available via both PostgreSQL and Athena
• CloudFront CDN: Global content delivery with HTTPS, caching, and custom error pages

Getting Started

Prerequisites

1. AWS CLI: Installed and configured with appropriate credentials.
2. PowerShell: Available on your system to run the scripts.
3. uv: Python package manager (installed by setup.ps1 if needed).

Quick Start

# 1. Initial setup
.\setup.ps1
.venv\Scripts\Activate.ps1

# 2. Deploy AgentCore supervisor
agentcore configure -e agent/supervisor.py
agentcore launch

# 3. Deploy infrastructure
.\deploy-phase1.ps1     # Web interface + chat handler
.\deploy-phase2.ps1     # Athena API + HDB database
.\deploy-phase3.ps1     # PostgreSQL API + RDS Proxy
.\deploy-cloudfront.ps1 # CloudFront distribution (global CDN)

# 4. Visit the CloudFront URL for global access

Key Features

✅ Cost-Optimized Multi-Agent System

• Property Search: Real HDB transactions via both PostgreSQL (fast) and Athena (complex)
• Cost Intelligence: Automatic routing for 70% cost savings on frequent queries
• Price Analytics: Statistics, trends, and market comparisons
• Demographics: Live Singapore government data via Data Commons API
• Intelligent Routing: Supervisor routes to optimal agent based on query complexity and cost

✅ Production Architecture

• AgentCore Runtime: AWS Bedrock AgentCore with Amazon Nova Lite model
• Dual Data Sources: PostgreSQL (RDS Proxy) + Athena for cost optimization
• RDS Proxy: Secure, scalable database connection pooling
• CloudFront CDN: Global content delivery with HTTPS and caching
• CloudFormation: Complete infrastructure as code
• Web Interface: Real-time chat interface with phase examples

🎯 Example Queries (Phase 3)

• "Average HDB prices in Tampines" → PostgreSQL Agent (cost-optimized, 70% cheaper)
• "Complex statistical analysis of 5-room price trends" → Property Agent (Athena for analytics)
• "Singapore population breakdown" → Demographics Agent (Data Commons API)

Design Principles

• AWS Official Pattern: Uses the supervisor-agent pattern from AWS multi-agent blog post
• Amazon Nova Lite: Cost-effective AI model for intelligent query routing and responses
• Specialized Agents: Each agent focuses on a specific domain (demographics, property data)
• Intelligent Routing: Supervisor routes queries to appropriate specialized agents
• Cost Optimization: Automatic selection between PostgreSQL (fast/cheap) and Athena (complex analytics)
• Standalone Architecture: No complex imports or inheritance - each agent is self-contained
• Memory-First: AgentCore Memory for conversation history and user preferences
• Scalable Foundation: Ready for AgentCore Gateway, Memory, and advanced multi-agent workflows