Building an Intelligent Multi-Agent AI Advisor for Property Investment: A Collaboration with PropHero

This heading emphasizes the collaborative effort and the innovative technology behind PropHero’s AI advisor, setting the stage for the detailed exploration that follows.

Transforming Property Investment with AI: An Inside Look at the PropHero Advisor

This post was crafted with insights from Lucas Dahan, Dil Dolkun, and Mathew Ng from PropHero.

Introduction

In today’s rapidly evolving real estate market, property investment remains a daunting task for many, whether they’re novices or veterans. Recognizing this challenge, PropHero, a pioneering property wealth management service, sought to democratize access to intelligent property investment advice using big data, AI, and machine learning (ML). This blog explores how PropHero built an AI-powered advisory system tailored specifically for its Spanish and Australian consumer base.

The Challenge: Making Property Investment Accessible

The process of property investment is fraught with barriers: information asymmetry, complex regulations, and manual, time-consuming procedures hinder both new and experienced investors. For PropHero, the challenge was to deliver accurate, actionable advice in Spanish, capable of holding rich, multi-turn conversations. The final goal? To accompany users through every phase of their investment journey, from initial inquiries to the final settlement.

Building the Solution: AWS Generative AI Innovation

PropHero partnered with the AWS Generative AI Innovation Center to implement a multi-agent advisory system leveraging AWS’s generative AI services for continuous evaluation. This AI advisor facilitates natural language interactions about investment strategies and delivers personalized recommendations grounded in PropHero’s extensive market knowledge.

System Architecture Overview

The solution architecture comprises four core layers, ensuring seamless functionality:

1. Data Foundation Layer: Responsible for the storage and retrieval infrastructure.
2. Multi-Agent AI Layer: It’s here that the core intelligence resides, powered by:
   • Amazon Bedrock: Foundation models that drive specialized agents.
   • LangGraph: Manages workflows and conversation states.
   • AWS Lambda: Executes multi-agent logic.
3. Continuous Evaluation Layer: Monitors and improves conversation quality with tools like Amazon CloudWatch and Amazon EventBridge.
4. Application and Integration Layer: Facilitates secure external communication via the Amazon API Gateway.

Multi-Agent AI Advisor Architecture

The intelligence behind the advisor is orchestrated through LangGraph, which operates in a single Lambda function. Each agent handles specific tasks to ensure optimized performance.

Agent Composition and Model Selection

A thorough model selection strategy guided our choices, evaluating response quality, latency, and cost-efficiency. Each agent utilizes the most appropriate Amazon Bedrock model, as demonstrated in the following table:

Sample Conversation Experience

The PropHero AI effortlessly handles natural conversations in Spanish, offering a glimpse into user interactions:

Usuario: “Hola, ¿qué es PropHero exactamente?”
Asistente: “¡Hola! PropHero es una plataforma que te permite crear y optimizar tu patrimonio inmobiliario…”

This fluency ensures users feel at ease and supported throughout their investment journey.

Integrated Continuous Evaluation System

Central to the success of the advisory system is its integrated continuous evaluation mechanism. This system employs metrics such as Context Relevance, Response Groundedness, and Agent Goal Accuracy to ensure quality monitoring in real time.

Real-Time Evaluation Workflow

Data from conversations automatically triggers a Lambda function for evaluation as conversations are stored in DynamoDB. Parallel processing ensures a seamless experience, with each conversation being assessed across multiple dimensions simultaneously.

Implementation Insights and Best Practices

Developed over six weeks, the project unveiled valuable insights:

• Model Selection: Balancing performance with cost by using appropriate models for different tasks.
• Chunking and Retrieval: Semantic chunking proved more efficient than other methods, optimizing retrieval speed and accuracy.
• Multilingual Capabilities: The system adeptly handles queries in both Spanish and English.

Business Impact

The AI advisor has delivered significant business benefits for PropHero, including:

• Enhanced Customer Engagement: A 90% goal accuracy rate means users receive relevant advice, with over 50% actively engaging with the AI.
• Operational Efficiency: Automated responses have reduced the customer service workload by 30%.
• Scalable Growth: The serverless architecture scales seamlessly with customer demand.
• Cost Optimization: A strategic model selection has led to a 60% reduction in AI costs.
• Consumer Base Expansion: Effective support in Spanish has facilitated entry into new markets.

Conclusion

The PropHero AI advisor exemplifies the transformative potential of AWS generative AI services in creating intelligent, context-aware conversational agents. By combining a modular agent architecture with robust evaluation systems, PropHero enhances customer engagement while ensuring accurate, relevant responses. This innovative approach not only addresses the immediate needs of property investors but also sets the stage for continuous improvement in AI capabilities.

For those interested in building effective multi-agent AI advisors, the PropHero journey offers critical insights and fosters ongoing evolution in conversational AI.

About the Authors:

• Adithya Suresh: Deep Learning Architect at AWS Generative AI Innovation Center.
• Lucas Dahan: Head of Data & AI at PropHero.
• Dil Dolkun: Data & AI Engineer at PropHero.
• Mathew Ng: Technical Lead at PropHero.

For more information on leveraging AWS for generative AI solutions, connect with your account team today!