Personalized Property Search

Transforming Property Searches with Generative AI

Leading player in the American real estate industry, with nationwide listing of properties on web platforms, wanted to be ahead of competitors by adopting Generative AI enabled approaches that serve customers with higher accuracy and beyond static filters based search.
Data preparation with quality assurance should support current static & natural language base search.


Increment in web traffic for property search


Increment in new property listing


Increment in property deals

About the business

The Client is a versatile real estate agency in the USA; serving clients in both city and suburban areas and dealing with both residential and commercial properties. They offer personalized assistance to clients looking to buy, sell, rent, or manage properties, focusing on understanding each client’s unique needs. They  prioritizes building strong relationships with clients by providing exceptional service, timely communication, and transparent guidance throughout the real estate process. Client retention is achieved through delivering on promises, maintaining professionalism, and staying in touch with clients beyond transactions, aiming to be their trusted partner for all real estate needs.

Business Requirements

  • As The Client keeps doing well in their work, they’ve realized they need to change with the times. They observed that dynamic needs of clients can not be served with current static filters to search properties. Clients are moving away from the old, complicated ways and making it easier and friendlier for people to search for properties. 
  • Static filters produce results; how listings are labelled or described in the database, leading to semantic gap, client’s need was to leverage machine learning techniques to continuously learn from user interactions and preferences. 
  • Improve filters behaviours which understand the internet behind user queries and matching with properties that closely align with their preferences, even if the terminology used varies.
  • Additionally, they want to implement a CRM system to keep proper records of their clients’ interactions and preferences. This helps them provide better service and solutions tailored to each client’s needs.

Real Estate

Headquarters in the New York

Property listing for sale, rent & lease
5,000 plus listed property on site.


  • Static filters restrict users to predefined search criteria, limiting their ability to express nuanced preferences.
  • Property databases may have incomplete or inaccurate information, leading to missed opportunities or irrelevant search results
  • Static filters typically provide a one-size-fits-all approach, lacking personalization for individual users’ preferences and behaviors.


AI Property Search Mockup

Transforming Property Search with Gen AI

The Solution

  • To meet this need, we’ve upgraded their property search system with advanced Generative AI technology based on LLM(Large Language Model). This new system makes things a lot simpler by getting rid of complicated filters. Instead, customers can just ask a question about what they’re looking for, and they’ll get accurate results that match their needs exactly. This new way of searching not only saves time but also makes the experience better by giving personalized recommendations.
  • In addition, we have implemented CRM for the Client to organize their customer’s information, enhancing service delivery. Centralized data enables efficient personalized assistance. CRM streamlines communication and automates tasks, improving team collaboration and customer satisfaction.

Customer Support Interface:

  • In parallel with upgrading their property search system, we’ve introduced an advanced customer support interface. This innovative solution, powered by cutting-edge Generative AI, allows customers to swiftly obtain assistance and answers to their queries. 
  • Our primary aim is to guarantee a seamless and satisfying experience for the customer throughout their interactions with our client, ensuring their needs are promptly addressed and resolved.


  • Generative AI algorithms can analyze and interpret user queries to understand underlying preferences, helping to retrieve relevant listings even if they don’t perfectly match the specified criteria.
  • Generative AI can employ machine learning techniques to continuously learn from user interactions and preferences, providing personalized recommendations and search results tailored to each user’s unique needs.
  • Generative AI can process complex queries by breaking them down into smaller components, understanding the relationships between different criteria, and generating optimized search results that balance conflicting preferences effectively.


Following is a Large Language Model-based property search algorithm for a leading property listing company that involves integrating various technologies to manage data ingestion, processing, model training, deployment, and user interaction.

  • Data Storage & Management

    • PostgreSQL: For storing structured property data, user profiles, and transaction records.

    • MongoDB: For storing unstructured data like user queries, logs, and feedback.
    • Elasticsearch: For fast search capabilities across vast datasets of property listings.

  • Data Ingestion & Processing

    • Apache Kafka: For real-time data ingestion from various sources like property listings, user interactions, and external APIs.

    • Apache Spark: For large-scale data processing, cleaning, and preparation before feeding it into the model.

  • Model Training & Machine Learning

    • PyTorch: A main framework for training the LLM on property data and user queries.

    • Hugging Face Transformers: For leveraging pre-trained models and fine-tuning them for property search-specific tasks.

    • MLflow: For managing the machine learning lifecycle, including experimentation, reproducibility, and deployment.

  • Pre-trained Models

    • Openai, Llama, Claude, Mistral, Mixtral, and Gemini Models for question-answering and text-generation tasks.

  • API & Microservices

    • FastAPI: For building fast, scalable APIs that serve the trained model to client applications.

    • Docker: For containerizing the application and its dependencies for easy deployment and scaling.

    • Kubernetes: For orchestrating and managing containers, ensuring high availability, and scaling the application based on demand.

  • User Interface & Frontend

    • React.js: For building a dynamic, responsive frontend for the web application.

    • React Native: For developing cross-platform mobile applications for iOS and Android.

  • Cloud User Interface & Frontend

    • AWS: For hosting the application, and databases, and for leveraging cloud-specific ML tools and services for training and inference.

    • Amazon S3: For storing large datasets and model artifacts.

    • Amazon EC2: For compute resources to train models.

    • AWS Lambda: For server-less functions to handle user queries and other back-end tasks.

  • Monitoring, Logging & Security

    • Prometheus: For monitoring the health and performance of the application and infrastructure.

    • Elasticsearch, Logstash, and Kibana (ELK Stack): For logging and visualizing logs from different parts of the application.

    • OAuth2.0: For securing APIs and protecting user data.

risk assessment tools
Find your dream property effortlessly? Try Generative AI-powered search now! 

Let’s connect and discuss how we can help bring your vision to life!

Send Message