LLM-Based Re-Ranking for Real Estate Search

2026-07-16

AI Summary

This work addresses the limitations of traditional property search systems, which rely on static filters and struggle to capture users’ context-dependent, multi-dimensional, and fine-grained intents expressed during multi-turn conversations. To overcome this, we propose the first integration of large language models (LLMs) into conversational property recommendation, leveraging LLMs to dynamically rerank candidate listings by synthesizing user intent from dialogue history. We construct a large-scale offline evaluation dataset comprising 960,000 query–property pairs and assess performance using an LLM-as-a-Judge framework validated with human evaluation. Both offline and online experiments demonstrate significant effectiveness: in production, our approach yields a 5.3% increase in click-through rate and a 4.8% rise in property viewing appointments.

Abstract

QuintoAndar Group operates the leading housing marketplace in Latin America for both rentals and sales. The platform replaces traditionally paper-heavy workflows with a fully digital experience, making housing transactions faster and more accessible to tenants, buyers, and landlords in the region. Finding the ideal home in such a vast catalog is inherently difficult. At the same time, the widespread adoption of conversational assistants is reshaping user expectations: people increasingly want to express their needs through open, multi-turn dialog rather than rigid filter menus and faceted search. This shift is particularly pronounced in housing, where intent is multi-dimensional, context-dependent, and rarely reducible to a small set of structured constraints. To meet these expectations, we propose a Large Language Model (LLM) based re-ranker that augments a conversational recommendation system by reordering retrieved candidates according to the nuanced, context-rich intent expressed across the user's conversation. We additionally construct a large-scale offline evaluation dataset for conversational real-estate search, containing 960,000 query-item pairs constructed from both synthetic and production queries and annotated using an LLM-as-a-Judge framework with human validation. We validate our approach both offline, on this proprietary dataset, and online, through a production A/B test. Both evaluations show consistent improvements in ranking quality, including a statistically significant increase in production of +5.3% in click-through rate and +4.8% in scheduled visits, demonstrating the value of integrating conversational context into housing recommendations.

Problem

• real estate search
• conversational recommendation
• user intent
• multi-turn dialogue
• housing marketplace

Innovation

• LLM-based re-ranking
• conversational search
• real estate recommendation
• LLM-as-a-Judge
• context-aware ranking