HabitatAgent: An End-to-End Multi-Agent System for Housing Consultation
HabitatAgent achieves 95% accuracy in housing consultation using a multi-agent system.
Key Findings
Methodology
HabitatAgent employs four specialized agents: Memory, Retrieval, Generation, and Validation. The Memory Agent updates user memory through verification-gated updates; the Retrieval Agent uses hybrid vector-graph retrieval; the Generation Agent provides evidence-backed recommendations; the Validation Agent performs multi-tier verification and targeted remediation.
Key Results
- HabitatAgent achieved 95% accuracy in 100 real consultation scenarios, significantly outperforming Dense+Rerank's 75%.
- For complex queries, HabitatAgent's CSR@5 reached 95%, while Dense+Rerank was only 8%.
- Removing adaptive retrieval routing reduced accuracy from 95% to 75%, highlighting its critical role.
Significance
HabitatAgent provides a reliable end-to-end solution for housing consultation, addressing existing platforms' shortcomings in multi-constraint handling and transparency, with significant academic and industrial impact.
Technical Contribution
HabitatAgent transforms housing consultation from mere recommendation to comprehensive decision support through a multi-agent architecture, introducing new mechanisms like verification-gated memory and adaptive retrieval routing.
Novelty
This is the first architecture combining LLM with a multi-agent system for housing consultation, significantly improving decision accuracy under multi-constraint conditions.
Limitations
- In extremely complex constraint conditions, the system may require longer processing times.
- Applicability to emerging markets remains to be verified.
Future Work
Future work will expand to more cities and dynamic knowledge graphs, enhancing system adaptability and real-time capabilities.
AI Executive Summary
Housing selection is a high-stakes and largely irreversible decision-making process, with existing platforms falling short in handling multi-constraints and transparency. HabitatAgent provides an end-to-end solution through a multi-agent system comprising Memory, Retrieval, Generation, and Validation agents. Experiments show that HabitatAgent achieves 95% accuracy in 100 real consultation scenarios, significantly outperforming existing methods. This system holds significant academic importance and offers a new decision support tool for the industry. Future work will expand to more cities and dynamic knowledge graphs, enhancing system adaptability and real-time capabilities.
Deep Analysis
Background
Housing selection involves multiple factors such as budget, location, and building quality. Existing platforms primarily focus on ranking or recommendation, lacking transparent decision support.
Core Problem
Existing platforms perform poorly in handling multi-constraint conditions, with opaque recommendation results that fail to meet personalized user needs.
Innovation
HabitatAgent transforms housing consultation from mere recommendation to comprehensive decision support through a multi-agent architecture, introducing new mechanisms like verification-gated memory and adaptive retrieval routing.
Methodology
- �� Memory Agent: Updates user memory through verification-gated updates. • Retrieval Agent: Uses hybrid vector-graph retrieval. • Generation Agent: Provides evidence-backed recommendations. • Validation Agent: Performs multi-tier verification and targeted remediation.
Experiments
Evaluated using real user interaction data from a Beijing housing platform, including 100 consultation scenarios and 300 user queries.
Results
HabitatAgent excels in complex queries, achieving a CSR@5 of 95%, significantly outperforming Dense+Rerank.
Applications
Applicable to real estate platforms, providing personalized housing recommendations and decision support.
Limitations & Outlook
In extremely complex constraint conditions, the system may require longer processing times.
Plain Language Accessible to non-experts
Imagine you're shopping in a large supermarket, and HabitatAgent is like a smart shopping assistant. It not only helps you find products that meet your requirements but also explains why these products suit you and ensures all information is accurate.
ELI14 Explained like you're 14
HabitatAgent is like a super-smart house assistant! It helps you find the best house and explains why it's the best choice. It's like having an awesome NPC in a game helping you make decisions, cool right?
Glossary
Multi-Agent System
A system that decomposes tasks into multiple specialized agents, each responsible for specific functions.
Used in HabitatAgent for housing consultation with four agents.
Verification-Gated Memory
A mechanism that only allows verified information to update user memory.
Prevents error propagation across multi-turn dialogues.
Adaptive Retrieval Routing
A mechanism that decides whether to use graph retrieval based on query complexity.
Preserves relational correctness in complex queries.
CSR@5 (Constraint Satisfaction Rate)
The proportion of recommended items in the top-5 that satisfy all hard constraints.
Used to evaluate HabitatAgent's performance in complex queries.
GraphRAG (Graph Retrieval-Augmented Generation)
A technique combining graph retrieval and generation to handle complex relational constraints.
Core technology used in HabitatAgent's Retrieval Agent.
Open Questions Unanswered questions from this research
- 1 How to maintain system real-time accuracy in dynamically changing markets?
- 2 Applicability and scalability in emerging markets?
Applications
Immediate Applications
Real Estate Platforms
Provides users with personalized housing recommendations and decision support, enhancing user satisfaction.
Long-term Vision
Smart City Planning
Integrates city data for intelligent urban planning and resource allocation.
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. The Memory Agent maintains multi-layer user memory through internal stages for constraint extraction, memory fusion, and verification-gated updates; the Retrieval Agent performs hybrid vector--graph retrieval (GraphRAG); the Generation Agent produces evidence-referenced recommendations and explanations; and the Validation Agent applies multi-tier verification and targeted remediation. Together, these agents provide an auditable and reliable workflow for end-to-end housing consultation. We evaluate HabitatAgent on 100 real user consultation scenarios (300 multi-turn question--answer pairs) under an end-to-end correctness protocol. A strong single-stage baseline (Dense+Rerank) achieves 75% accuracy, while HabitatAgent reaches 95%.