Modernizing Real Estate Operations With Agentic AI

Overview

A prominent real estate client partnered with JBS Dev to create an AI-powered solution designed to modernize and simplify the creation of MLS home listings. In an increasingly competitive real estate market, compounded by recent commission rule changes, realtors needed to reduce administrative workload and focus more time on selling activities.

Existing MLS platforms relied heavily on manual data entry and outdated workflows, resulting in slow, error-prone listing creation and making it difficult to scale. The client sought an automated, intelligent solution that could accelerate listing creation while improving accuracy and consistency.

Challenges

Realtors faced time-consuming, manual processes when creating and submitting MLS listings, often spending hours, or even an entire day, per listing. These workflows increased the likelihood of inaccuracies, exposing realtors to potential compliance issues and financial penalties.

Additionally, MLS platforms were outdated and inflexible, with limited integration support. This lack of extensibility forced realtors to manually re-enter data across systems, slowing time-to-market for new listings and negatively impacting both competitiveness and client satisfaction.

Solution

JBS Dev designed and delivered a comprehensive, AI-driven platform built on a modern AWS-based architecture. The solution focused on automating listing creation by extracting and generating high-quality, structured listing data from images and supporting content.

Key components included:

• AI-Powered Agents
  Using AWS S3 and Amazon Bedrock, integrated through the LangChain framework, AI agents analyze listing photos and text to generate rich descriptions and identify notable property features. Outputs are produced as structured JSON, enabling consistency and downstream automation.
• Modern Web Portal
  A Django-based web application hosted on AWS Lambda with AWS RDS as the backend datastore allows realtors to upload photos, manage listings, and review or edit AI-generated content before submission.
• MLS System Integration
  Due to the lack of native MLS APIs, our team used a Chrome browser extension to automate form population directly within MLS systems, eliminating repetitive manual entry.
  

How AWS Was Used to Solve the AI Challenge

A core advantage of the solution was the use of Amazon Bedrock, which provided exceptional flexibility in model selection. The team was able to test and switch between multiple foundation models, including different versions of Claude and OpenAI models, directly through an administrative interface, without requiring changes to application code.

This flexibility allowed rapid experimentation with reasoning styles and output quality, helping the team identify which models produced the most accurate and contextually appropriate listing descriptions. As new models became available, the system could evolve without disrupting the underlying architecture.

Combined with a serverless foundation built on AWS Lambda and AWS CDK, S3, and RDS, Bedrock enabled highly adaptable, efficient, scalable, and future-proof agentic workflows.

Lessons Learned

Model and Prompt Flexibility Were Critical

Prompt design had a significant impact on output quality. Even minor changes produced noticeable differences in listing accuracy and tone. To support continuous improvement, JBS Dev built an administrative interface that allowed non-developers to modify prompts in real time. This eliminated deployment cycles and enabled rapid experimentation driven by business users and realtors.

Keeping Context Small Reduced Hallucinations

Large, monolithic prompts tended to increase hallucinations and reduce reliability. The team addressed this by breaking each listing into atomic components:

• Each image is processed independently
• Home features are extracted as discrete structured elements
• Descriptions are generated in focused, isolated steps

By assembling final descriptions only after validating each component, the system consistently produced high-precision, factual outputs.

Rapid Feedback Loops Improved Output Quality

With both model selection and prompt configuration available without code changes, iteration cycles were extremely tight. Realtors and internal testers could compare outputs, provide immediate feedback, and quickly converge on results that matched real-world expectations, keeping AI behavior aligned with user needs.

Results

The platform reduced MLS listing creation time from up to a full day to less than 10 minutes. AI-generated descriptions were richer, more accurate, and more engaging, driven by detailed feature recognition from listing photos.

Realtors experienced significant gains in productivity, reduced administrative burden, and improved listing accuracy, allowing them to focus more time on sales and client relationships while accelerating time-to-market for new properties.

Future Initiatives

AI-Driven Buyer–Property Matchmaking

A major upcoming enhancement is the introduction of an AI-powered buyer–property matchmaking engine. Leveraging the platform’s increasingly rich, structured home data, the system will analyze buyer profiles, including preferred features, locations, styles, and budget ranges, and automatically match them with relevant listings.

This capability will enable the platform to:

• Identify properties aligned with explicit buyer preferences
• Surface recommendations based on deeper behavioral patterns
• Highlight unique home attributes relevant to specific buyer segments
• Provide realtors with prioritized match lists for faster outreach

This evolution positions the platform not only as a listing automation tool, but as an intelligent advisor supporting the entire buyer journey, from discovery through engagement.