Introduction
Property management is a data‑intensive field that relies on quick access to a vast array of documents, contracts, leases, inspection reports, and regulatory filings. Traditionally, professionals in the industry have faced fragmented systems, manual searches, and cumbersome data pipelines that slow decision‑making and increase operational risk. In a recent collaboration, CBRE, the world’s largest commercial real‑estate services firm, partnered with Amazon Web Services to build a unified search platform and a conversational digital assistant powered by Amazon Bedrock. By integrating Bedrock’s foundation models—Claude Haiku for natural‑language document interaction and Amazon Nova Pro for SQL generation—CBRE was able to provide a single, secure interface that lets users query millions of documents and multiple databases with plain English. The result was a dramatic 67 % reduction in processing time while maintaining enterprise‑grade security across more than eight million documents. This post explores the architecture, the technology choices, the security posture, and the tangible benefits that emerged from this partnership.
Main Content
Unified Search Architecture
At the heart of the solution lies a two‑tier architecture that separates data ingestion from query execution. The ingestion layer pulls data from heterogeneous sources—structured relational databases, semi‑structured JSON files, and unstructured PDFs—into a consolidated data lake on Amazon S3. Metadata extraction is performed using Amazon Textract and AWS Glue, which normalizes the data into a common schema. The search layer leverages Amazon OpenSearch Service, where each document is indexed with rich semantic embeddings generated by Bedrock’s Claude Haiku model. These embeddings enable similarity‑based retrieval, allowing the system to surface the most contextually relevant documents even when the query does not match exact keywords.
The search layer is exposed through a GraphQL API that abstracts the underlying complexity. When a user submits a natural‑language query, the API first passes the text to Claude Haiku, which interprets intent and extracts entities such as property identifiers, dates, or regulatory terms. The model then constructs a structured query that is sent to Amazon Nova Pro, which translates the intent into a precise SQL statement. Nova Pro’s ability to generate syntactically correct SQL on the fly eliminates the need for manual query writing, thereby reducing the cognitive load on property managers.
Natural Language Interaction
Claude Haiku is a lightweight, yet powerful, large language model that excels at conversational tasks. In this deployment, it serves two primary functions. First, it acts as a semantic layer that maps user intent to database schema elements. For example, a user asking “Show me the lease renewal status for the 5th floor of the downtown office tower” is parsed into a structured request that references the correct table, column, and filter conditions. Second, it provides a document‑centric conversational interface. Users can ask follow‑up questions such as “What were the key clauses in the last renewal?” and Claude Haiku can retrieve the relevant paragraph from the PDF, summarize it, or even generate a concise bullet list—all while preserving the original document’s context.
The conversational flow is designed to mimic a human assistant. The system maintains a short‑term memory of the conversation, allowing it to answer follow‑up queries without requiring the user to restate context. This feature is particularly valuable in property management, where a single project may involve dozens of documents and stakeholders.
Security and Compliance
Handling sensitive real‑estate data demands rigorous security controls. CBRE’s implementation leverages AWS’s built‑in compliance frameworks, including ISO 27001, SOC 2, and GDPR. All data at rest is encrypted using AWS Key Management Service (KMS) with customer‑managed keys. In transit, TLS 1.3 is enforced across all API endpoints. Access to the data lake and search indices is governed by fine‑grained IAM policies that enforce least‑privilege principles.
To address the risk of model hallucination—a common concern with generative AI—CBRE instituted a verification layer. Claude Haiku’s responses are cross‑checked against the source documents before being presented to the user. If the model attempts to generate information not present in the data set, the system flags the response and prompts the user to review the original source. This approach preserves the integrity of the information while still delivering the convenience of a conversational assistant.
Performance Gains
The most striking metric from the pilot was the 67 % reduction in processing time for document retrieval and query execution. Prior to the Bedrock‑powered solution, property managers spent an average of 12 minutes per search session, juggling multiple tools and manual filtering. After deployment, the average time dropped to just 4 minutes. This improvement was achieved through several optimizations: vector‑based similarity search in OpenSearch, pre‑computed embeddings, and the elimination of manual SQL writing thanks to Nova Pro’s instant query generation.
Beyond speed, the unified platform also reduced the number of support tickets related to data access. Because the system presents a single, consistent interface, users no longer need to consult IT for database credentials or learn new query languages. This shift freed up IT staff to focus on higher‑value tasks such as data governance and model fine‑tuning.
Future Outlook
CBRE is already exploring extensions to the platform. One avenue is the integration of multimodal models that can interpret images and floor plans, enabling visual search capabilities. Another is the deployment of a reinforcement‑learning loop where user interactions are logged to continuously improve Claude Haiku’s intent‑recognition accuracy. Finally, CBRE plans to expose the API to third‑party partners, creating an ecosystem where external vendors can query property data securely and efficiently.
Conclusion
The partnership between CBRE and AWS demonstrates how generative AI, when thoughtfully integrated with robust data pipelines and security controls, can transform a traditionally fragmented industry. By leveraging Amazon Bedrock’s Claude Haiku and Nova Pro, CBRE was able to create a unified search and digital assistant that not only speeds up information retrieval but also maintains stringent compliance standards. The 67 % reduction in processing time is a tangible testament to the power of AI‑driven automation in real‑estate operations. As the technology matures, we can expect even deeper integrations, richer conversational experiences, and broader adoption across the property management landscape.
Call to Action
If you’re a property manager, data engineer, or AI enthusiast looking to streamline your organization’s data workflows, consider exploring Amazon Bedrock’s foundation models. Reach out to your AWS account team to discuss how Claude Haiku and Nova Pro can be tailored to your specific use case. By embracing these tools, you can unlock faster insights, reduce operational friction, and position your business at the forefront of AI‑enabled real‑estate management. Take the first step today and discover how unified search and conversational AI can elevate your property portfolio to new heights.