This post was written with Lucas Dahan, Dil Dolkun, and Mathew Ng from PropHero.
PropHero is a leading property wealth management service focused on democratizing access to intelligent property investment advice through big data, AI, and machine learning (ML). Recognizing the needs of their Spanish and Australian consumer base, PropHero sought an AI-powered advisory system capable of engaging users in accurate discussions related to real estate. The objective was to deliver personalized investment insights and guide users throughout their journey—from understanding the process to securely uploading documents and tracking progress in real time.
PropHero partnered with the AWS Generative AI Innovation Center to implement this intelligent advisor, leveraging AWS generative AI services alongside a continuous evaluation system. This solution allows users to participate in natural language conversations about property investment strategies and receive tailored recommendations based on PropHero’s extensive market knowledge.
In this post, we will explore how we built a multi-agent conversational AI system using Amazon Bedrock to provide knowledgeable and grounded advice regarding property investment. We’ll examine the agent architecture, model selection strategy, and comprehensive evaluation system that facilitates high-quality conversations while allowing for rapid iteration and improvement.
Understanding the Challenges in Property Investment
The world of property investment presents significant challenges for investors of all levels. Information asymmetry often creates barriers due to the cost or inaccessibility of comprehensive market data. Furthermore, traditional processes are frequently manual and time-consuming, demanding extensive market knowledge to navigate effectively. For PropHero’s Spanish and Australian consumers, building a solution that provides accurate, contextually relevant advice while handling complex conversations was essential. The system needed to maintain high accuracy at scale, continuously learning and improving from user interactions. Crucially, it had to support users throughout every phase of their investment journey, ensuring comprehensive assistance.
Solution Overview: A Multi-Agent Approach
We developed a complete end-to-end solution leveraging AWS generative AI services, centered around a multi-agent AI advisor with integrated continuous evaluation. This system ensures seamless data flow from ingestion to intelligent advisory conversations, coupled with real-time quality monitoring. The architecture illustrated below provides a visual representation of this process.
The solution architecture comprises four virtual layers, each fulfilling specific functions within the overall system design. Initially, a data foundation layer serves as the bedrock of the entire infrastructure.
Data Foundation Layer: The Source of Truth
This layer provides essential storage and retrieval capabilities:
- Amazon DynamoDB is used for fast storage of conversation history, evaluation metrics, and user interaction data.
- Amazon Relational Database Service (RDS) for PostgreSQL stores comprehensive property data, market trends, and investment guidelines.
Agent Orchestration Layer: Managing Conversational Flow
The agent orchestration layer leverages Amazon Bedrock to manage the interaction between various specialized agents. These include a User Intent Agent to determine user needs and guide conversation flow, a Knowledge Retrieval Agent to access relevant data from DynamoDB and RDS, and a Response Generation Agent responsible for crafting natural language responses based on retrieved information.
Evaluation Layer: Ensuring Quality and Continuous Improvement
This critical layer continuously monitors the quality of conversations, providing valuable feedback for iterative improvement. Automated metrics, such as accuracy and relevance, are calculated using Amazon Bedrock’s assessment models. Furthermore, human reviewers provide qualitative insights into conversation performance.
Presentation Layer: User Interface & Real-Time Tracking
The presentation layer delivers a user-friendly interface for interacting with the AI advisor. Users can engage in natural language conversations through web and mobile applications. In addition, real-time progress tracking visually represents investment timelines and milestones, providing users with clear visibility into their property investment journey.
Conclusion: Democratizing Property Investment
By leveraging Amazon Bedrock and a comprehensive evaluation system, PropHero has created a powerful tool to democratize access to intelligent property investment advice. This innovative approach not only simplifies complex processes but also empowers users with the knowledge they need to make informed decisions throughout their investment journey.
Source: Read the original article here.
Discover more from ByteTrending
Subscribe to get the latest posts sent to your email.
