HabitatAgent: An End-to-End Multi-Agent System for Housing Consultation
Summary: arXiv:2604.00556v1 Announce Type: cross
Abstract
Housing selection is a high-stakes and largely irreversible decision problem. We study housing consultation as a decision-support interface for housing selection. Existing housing platforms and many LLM-based assistants often reduce this process to ranking or recommendation, resulting in opaque reasoning, brittle multi-constraint handling, and limited guarantees on factuality.
We present HabitatAgent, the first LLM-powered multi-agent architecture for end-to-end housing consultation. HabitatAgent comprises four specialized agent roles: Memory, Retrieval, Generation, and Validation.
Agent Roles in HabitatAgent
The HabitatAgent system is designed around four specialized agents, each fulfilling a crucial role in the housing consultation process:
- Memory Agent: This agent maintains multi-layer user memory through internal stages for constraint extraction, memory fusion, and verification-gated updates.
- Retrieval Agent: The Retrieval Agent performs hybrid vector-graph retrieval, known as GraphRAG, to efficiently source pertinent information.
- Generation Agent: This agent is responsible for producing evidence-referenced recommendations and explanations tailored to user queries.
- Validation Agent: The Validation Agent ensures the reliability of the information provided by applying multi-tier verification and targeted remediation.
Benefits of Using HabitatAgent
By integrating these specialized agents, HabitatAgent offers an auditable and reliable workflow for end-to-end housing consultation. The system addresses common challenges faced by traditional housing platforms:
- Opacity in Reasoning: Unlike conventional systems that offer mere rankings, HabitatAgent provides comprehensive explanations for its recommendations.
- Robust Multi-Constraint Handling: The Memory Agent’s capabilities allow for better management of user constraints, ensuring that suggestions align with user needs.
- Factual Accuracy: The Validation Agent’s multi-tier verification enhances the factuality of the recommendations, providing users with more trustworthy options.
Performance Evaluation
We evaluated HabitatAgent on 100 real user consultation scenarios, which included 300 multi-turn question-answer pairs. The evaluation was conducted under an end-to-end correctness protocol to assess its effectiveness in providing accurate housing consultations.
A strong single-stage baseline system, known as Dense+Rerank, achieved an accuracy of 75%. In contrast, HabitatAgent demonstrated a remarkable accuracy of 95%, showcasing its significant improvement over existing solutions.
Conclusion
HabitatAgent represents a significant advancement in the landscape of housing consultation. By leveraging a multi-agent architecture powered by LLM technology, it not only enhances decision-making efficiency but also ensures transparency and accuracy in the housing selection process. As housing decisions carry substantial weight, systems like HabitatAgent are poised to transform how individuals navigate these critical choices.