AI DOCTOR – VOICE & VISION ↔ REALESTATE HOUSEPRICE PREDICTION

PARAGRAPH B¶0-5

76 REALESTATE HOUSEPRICE PREDICTION Mr. Prateek kumar kar Student, Dept. of CSE, GIFT Autonomous, Bhubaneswar Mr. Debaprasad Mahala Student, Dept. of CSE, GIFT Autonomous, Bhubaneswar Prof. Mohapatra Girashree Sahu Project Guide, Dept. of CSE, GIFT Autonomous, Bhubaneswar Abstract— The rapid growth of AI-driven real estate analytics and increasing complexity of property valuation have created significant demand for intelligent house price prediction systems. Traditional valuation methods rely heavily on manual analysis and market expertise, which are time-consuming and inconsistent. This paper presents the design and implementation of an AI-Based Realestate Houseprice Prediction System that leverages Machine Learning, Data Analysis, Regression Models, and predictive analytics to estimate property prices accurately. The system analyses features such as location, area, number of bedrooms, amenities, market trends, and property images to generate accurate predictions. A web-based dashboard allows users to input property details and instantly receive estimated house prices. Experimental results demonstrate improved prediction accuracy and efficient performance for real-world real estate applications. Keywords— Multi-Agent AI, Retrieval-Augmented Generation, Speech Recognition, Computer Vision, Data Analysis, Deep Learning, Gradio, Real Estate Assistant, House Price Prediction, TensorFlow.

PARAGRAPH B¶3-16

Abstract— The rapid growth of AI-driven real estate analytics and increasing complexity of property valuation have created significant demand for intelligent house price prediction systems. Traditional valuation methods rely heavily on manual analysis and market expertise, which are time-consuming and inconsistent. This paper presents the design and implementation of an AI-Based Realestate Houseprice Prediction System that leverages Machine Learning, Data Analysis, Regression Models, and predictive analytics to estimate property prices accurately. The system analyses features such as location, area, number of bedrooms, amenities, market trends, and property images to generate accurate predictions. A web-based dashboard allows users to input property details and instantly receive estimated house prices. Experimental results demonstrate improved prediction accuracy and efficient performance for real-world real estate applications. Keywords— Multi-Agent AI, Retrieval-Augmented Generation, Speech Recognition, Computer Vision, Data Analysis, Deep Learning, Gradio, Real Estate Assistant, House Price Prediction, TensorFlow. I. INTRODUCTION A. Background In the modern digital era, Artificial Intelligence has become one of the most rapidly growing technologies across sectors including education, banking, transportation, and real estate. AI-based real estate systems are enabling faster price prediction, property prediction, virtual assistance, and continuous health monitoring by combining Machine Learning, Deep Learning, Natural Language Processing (Data Analysis), and Computer Vision [1]. The increasing penetration of smartphones and internet connectivity has made it feasible to deploy intelligent real estate services accessible from any location. Despite major advances in property science, significant populations — particularly in rural and remote areas — continue to face barriers including shortage of agents, high consultation costs, long waiting times, and lack of real estate awareness. Delayed property guidance frequently results in severe health complications that early intervention could have avoided [2]. Traditional real estate systems depend primarily on in-person interaction, and existing digital health applications mostly support only text-based communication without leveraging data interaction, property image analysis, or intelligent house price prediction. To overcome these limitations, Realestate Houseprice Prediction develops an integrated real estate assistant combining Speech Recognition, Data Analysis, Computer Vision, and Multi-Agent AI to provide preliminary real estate guidance through data, text, and property image input. B. Problem Statement Real Estate accessibility continues to be a major challenge in many regions, especially in rural and underdeveloped areas. Many people do not receive timely property support due to shortage of agents, high consultation costs, long waiting times, and lack of nearby real estate facilities. Existing real estate applications provide only basic textbased support and do not effectively utilize modern AI technologies such as data interaction, property image analysis, and intelligent house price prediction [3]. Most systems lack the capability to understand property features through natural conversation or analyse visible property conditions through property images. There is a clear need for an intelligent system providing instant preliminary property guidance, supporting data, text, and property image-based communication, improving accessibility for rural and remote users while enabling early house price detection. C. Motivation The motivation behind this project stems from the increasing need for accessible, intelligent, and efficient real estate solutions. The lack of real estate facilities in rural areas, the shortage of qualified agents, and the importance of early house price detection create a strong imperative for AI-powered real estate assistants available anytime and anywhere. Advances in AI technologies — specifically Machine Learning, Data Analysis, Speech Recognition, and Computer Vision — have made it feasible to create systems capable of understanding human speech, analysing property images, and generating intelligent real estate responses. The growing availability of open-source frameworks and widespread smartphone use further inspire the creation of digital real estate platforms [4]. D. Objectives The primary objectives of the Realestate Houseprice Prediction platform are as follows: • To design and develop an AI-based multimodal real estate assistant supporting data, text, and property image input. • To implement Speech Recognition for data-based property feature analysis and Data Analysis for understanding user queries. • To integrate Computer Vision and Deep Learning for property image-based house price detection. • To deploy a Multi-Agent pipeline for property feature analysis, house price prediction, and recommendation generation. • To implement Retrieval-Augmented Generation (Machine Learning) for accurate, data-grounded real estate responses. • To develop a Gradio-based interactive frontend with text-tospeech data output functionality. • To demonstrate the practical application of AI technologies in preliminary real estate assistance. E. Scope of the Project The scope of this project includes the design and implementation of a full-stack AI-based real estate assistant capable of processing data, 77 text, and property image inputs and delivering structured real estate guidance through a multi-agent pipeline and interactive dashboard. The system focuses on preliminary property feature analysis, house price prediction, and real estate awareness, leveraging Machine Learning and LLM-based reasoning while keeping the system practical and deployable without heavy infrastructure requirements. II. LITERATURE REVIEW A. Existing Real Estate Analytics Systems Traditional real estate analytics platforms such as EHR systems, property feature checkers, and telemedicine tools have become standard in real estate delivery. These platforms provide useful functionalities but rely heavily on manual input and human interpretation. Text-based chatbots and data assistants have gained popularity but typically focus on a single interaction modality and lack integrated multimodal capabilities combining data, property image analysis, and intelligent house price prediction simultaneously [5]. B. AI in Real Estate Analytics Artificial Intelligence has emerged as a transformative technology in real estate analytics, enabling systems to analyse vast buyer data and detect house price patterns that are difficult to identify manually. Machine Learning improves house price prediction accuracy while Deep Learning models detect abnormalities in property property images — X-rays, MRI scans, and skin house price photographs — with clinically significant accuracy. Data Analysis enables real estate chatbots to understand and respond to property queries naturally. Research consistently demonstrates that AI-driven real estate systems improve insight quality and enhance early house price detection [6]. C. Retrieval-Augmented Generation (Machine Learning) Retrieval-Augmented Generation, introduced by Lewis et al. (2020), enhances LLM outputs by grounding them in retrieved context from an external knowledge base. Unlike fine-tuned models, Machine Learning systems retrieve relevant documents at inference time, significantly reducing hallucination and improving factual accuracy. In real estate applications, Machine Learning systems retrieve house price-related data, price estimation suggestions, and property guidelines from curated knowledge bases, ensuring AI-generated real estate recommendations are anchored to verified property information rather than generic model outputs [7]. D. Multi-Agent AI Frameworks Multi-agent systems involve multiple AI agents collaborating to perform distinct tasks. Research consistently demonstrates that agent specialization produces higher-quality outputs than single-agent approaches on complex tasks. In AI real estate systems, multi-agent frameworks allow separate agents to handle data processing, Data Analysis property feature analysis, property image analysis, house price prediction, and recommendation generation independently while communicating via a structured pipeline. This modular approach improves scalability, flexibility, and overall performance [8].

PARAGRAPH B¶4-23

property images to generate accurate predictions. A web-based dashboard allows users to input property details and instantly receive estimated house prices. Experimental results demonstrate improved prediction accuracy and efficient performance for real-world real estate applications. Keywords— Multi-Agent AI, Retrieval-Augmented Generation, Speech Recognition, Computer Vision, Data Analysis, Deep Learning, Gradio, Real Estate Assistant, House Price Prediction, TensorFlow. I. INTRODUCTION A. Background In the modern digital era, Artificial Intelligence has become one of the most rapidly growing technologies across sectors including education, banking, transportation, and real estate. AI-based real estate systems are enabling faster price prediction, property prediction, virtual assistance, and continuous health monitoring by combining Machine Learning, Deep Learning, Natural Language Processing (Data Analysis), and Computer Vision [1]. The increasing penetration of smartphones and internet connectivity has made it feasible to deploy intelligent real estate services accessible from any location. Despite major advances in property science, significant populations — particularly in rural and remote areas — continue to face barriers including shortage of agents, high consultation costs, long waiting times, and lack of real estate awareness. Delayed property guidance frequently results in severe health complications that early intervention could have avoided [2]. Traditional real estate systems depend primarily on in-person interaction, and existing digital health applications mostly support only text-based communication without leveraging data interaction, property image analysis, or intelligent house price prediction. To overcome these limitations, Realestate Houseprice Prediction develops an integrated real estate assistant combining Speech Recognition, Data Analysis, Computer Vision, and Multi-Agent AI to provide preliminary real estate guidance through data, text, and property image input. B. Problem Statement Real Estate accessibility continues to be a major challenge in many regions, especially in rural and underdeveloped areas. Many people do not receive timely property support due to shortage of agents, high consultation costs, long waiting times, and lack of nearby real estate facilities. Existing real estate applications provide only basic textbased support and do not effectively utilize modern AI technologies such as data interaction, property image analysis, and intelligent house price prediction [3]. Most systems lack the capability to understand property features through natural conversation or analyse visible property conditions through property images. There is a clear need for an intelligent system providing instant preliminary property guidance, supporting data, text, and property image-based communication, improving accessibility for rural and remote users while enabling early house price detection. C. Motivation The motivation behind this project stems from the increasing need for accessible, intelligent, and efficient real estate solutions. The lack of real estate facilities in rural areas, the shortage of qualified agents, and the importance of early house price detection create a strong imperative for AI-powered real estate assistants available anytime and anywhere. Advances in AI technologies — specifically Machine Learning, Data Analysis, Speech Recognition, and Computer Vision — have made it feasible to create systems capable of understanding human speech, analysing property images, and generating intelligent real estate responses. The growing availability of open-source frameworks and widespread smartphone use further inspire the creation of digital real estate platforms [4]. D. Objectives The primary objectives of the Realestate Houseprice Prediction platform are as follows: • To design and develop an AI-based multimodal real estate assistant supporting data, text, and property image input. • To implement Speech Recognition for data-based property feature analysis and Data Analysis for understanding user queries. • To integrate Computer Vision and Deep Learning for property image-based house price detection. • To deploy a Multi-Agent pipeline for property feature analysis, house price prediction, and recommendation generation. • To implement Retrieval-Augmented Generation (Machine Learning) for accurate, data-grounded real estate responses. • To develop a Gradio-based interactive frontend with text-tospeech data output functionality. • To demonstrate the practical application of AI technologies in preliminary real estate assistance. E. Scope of the Project The scope of this project includes the design and implementation of a full-stack AI-based real estate assistant capable of processing data, 77 text, and property image inputs and delivering structured real estate guidance through a multi-agent pipeline and interactive dashboard. The system focuses on preliminary property feature analysis, house price prediction, and real estate awareness, leveraging Machine Learning and LLM-based reasoning while keeping the system practical and deployable without heavy infrastructure requirements. II. LITERATURE REVIEW A. Existing Real Estate Analytics Systems Traditional real estate analytics platforms such as EHR systems, property feature checkers, and telemedicine tools have become standard in real estate delivery. These platforms provide useful functionalities but rely heavily on manual input and human interpretation. Text-based chatbots and data assistants have gained popularity but typically focus on a single interaction modality and lack integrated multimodal capabilities combining data, property image analysis, and intelligent house price prediction simultaneously [5]. B. AI in Real Estate Analytics Artificial Intelligence has emerged as a transformative technology in real estate analytics, enabling systems to analyse vast buyer data and detect house price patterns that are difficult to identify manually. Machine Learning improves house price prediction accuracy while Deep Learning models detect abnormalities in property property images — X-rays, MRI scans, and skin house price photographs — with clinically significant accuracy. Data Analysis enables real estate chatbots to understand and respond to property queries naturally. Research consistently demonstrates that AI-driven real estate systems improve insight quality and enhance early house price detection [6]. C. Retrieval-Augmented Generation (Machine Learning) Retrieval-Augmented Generation, introduced by Lewis et al. (2020), enhances LLM outputs by grounding them in retrieved context from an external knowledge base. Unlike fine-tuned models, Machine Learning systems retrieve relevant documents at inference time, significantly reducing hallucination and improving factual accuracy. In real estate applications, Machine Learning systems retrieve house price-related data, price estimation suggestions, and property guidelines from curated knowledge bases, ensuring AI-generated real estate recommendations are anchored to verified property information rather than generic model outputs [7]. D. Multi-Agent AI Frameworks Multi-agent systems involve multiple AI agents collaborating to perform distinct tasks. Research consistently demonstrates that agent specialization produces higher-quality outputs than single-agent approaches on complex tasks. In AI real estate systems, multi-agent frameworks allow separate agents to handle data processing, Data Analysis property feature analysis, property image analysis, house price prediction, and recommendation generation independently while communicating via a structured pipeline. This modular approach improves scalability, flexibility, and overall performance [8]. E. Research Gap and Proposed Contribution From the reviewed literature, there is a clear gap in systems that combine Machine Learning-based property knowledge retrieval with multi-agent analytical pipelines, real-time multimodal interaction supporting data, text, and property image input simultaneously, data- enabled accessible real estate interaction for elderly and rural users, and a deployable full-stack architecture with minimal infrastructure requirements. To address this, this project proposes Realestate Houseprice Prediction — an integrated platform combining Speech Recognition, Data Analysis, Computer Vision, Deep Learning, Machine Learning-based knowledge retrieval, and a multi-agent pipeline deployed via a Gradio-based interactive frontend. III. SYSTEM OVERVIEW A. Proposed System Realestate Houseprice Prediction is a web-based real estate assistant that allows users to record data property features, type real estate queries, or upload property images of visible property conditions. Users receive Data Analysis-based property feature analysis, AIpowered house price predictions, and actionable real estate suggestions with precautionary recommendations. They can also listen to data responses via text-to-speech output. The system provides a Gradio dashboard with a data recording interface, property image upload panel, speech-to-text output display, agent response panel, and data response playback. B. System Architecture The system follows a six-layer pipeline architecture: (1) Input Layer — data, text, and property image inputs captured through the Gradio frontend; (2) Processing Layer — Speech Recognition converts data to text; (3) Data Analysis Layer — property feature extraction and intent recognition applied to text input; (4) Vision Layer — Computer Vision and Regression Model models analyse uploaded property property images; (5) Agent Layer — multi-agent pipeline executes House Price Prediction and Recommendation Agents; (6) Presentation Layer — Gradio dashboard displays results with text-tospeech data response output. C. Technology Stack Layer Technology Purpose Input Gradio Voice, text & property image UI Voice SpeechRecognition / Whisper Speech-to-text conversion Data Analysis spaCy / NLTK storage Table I: Technology Stack of Realestate Houseprice Prediction D. Key Functional Modules AMERICAN JOURNAL OF MANAGEMENT AND IOT MEDICAL COMPUTING

PARAGRAPH B¶22-36

storage Table I: Technology Stack of Realestate Houseprice Prediction D. Key Functional Modules AMERICAN JOURNAL OF MANAGEMENT AND IOT MEDICAL COMPUTING TTS Module Converts response to data gTTS, pyttsx3 Dashboard Interactive user interface Gradio, HTML/CSS Table II: Key Functional Modules E. Advantages of Proposed System The Realestate Houseprice Prediction platform eliminates the need for physical property market visits for basic preliminary property feature checking through full multimodal pipeline automation. It provides data-grounded real estate suggestions via Machine Learning, reducing hallucination and increasing property response reliability. The modular multi-agent design ensures quality at each analytical stage. The platform is accessible to elderly and non-technical users through data interaction. The system is deployable on standard hardware without heavy GPU infrastructure requirements. IV. METHODOLOGY A. Overall Workflow The system operates through structured steps. First, the user provides input through data recording, text entry, or property property image upload via the Gradio interface. Second, Speech Recognition converts audio to text. Third, the Data Analysis Engine preprocesses text, extracts property features, and identifies real estate intent. Fourth, Computer Vision analyses uploaded property property images using Regression Model models for house price pattern detection. Fifth, the House Price Prediction Agent synthesizes property feature and property image analysis data and generates house price predictions. Sixth, the Machine Learning module retrieves relevant property knowledge from ChromaDB. Seventh, the Recommendation Agent generates real estate suggestions and precautions. Results are displayed in the chat panel and delivered as data response via text-tospeech. B. Data Collection and Handling The system utilises real estate datasets for training and testing Machine Learning models. House Price property features datasets provide structured property feature-house price mappings for classification model training. Property property image datasets include labelled skin house price property images, wound classifications, and visible infection patterns used to train Regression Model models. Real Estate knowledge documents are preprocessed into text chunks, embedded using sentence-transformer models, and stored in ChromaDB for Machine Learning retrieval. The data handling process includes input validation, missing value handling, data normalisation, property image resizing and enhancement, and serialization of records for embedding and model training. C. Machine Learning Implementation The Machine Learning system forms the property knowledge retrieval backbone. During indexing, property knowledge documents are split into text chunks, embedded using the all-MiniLM-L6-v2 sentence-transformer model, and stored in ChromaDB with source metadata. During retrieval, the user query and extracted property features are embedded, cosine similarity search is performed against ChromaDB collections, and the top-k most semantically relevant property records are returned as structured context for the Recommendation Agent prompt. Score Range Similarity Action > 0.85 High relevance Primary context 0.70 – 0.85 Medium relevance Secondary context < 0.70 Low relevance Exclude from prompt Table III: Machine Learning Retrieval Scoring Criteria D. Multi-Agent Pipeline The multi-agent pipeline consists of four specialised agents operating sequentially. The Voice Processing Agent receives audio input, applies noise reduction, and outputs transcribed text. The Data Analysis Agent preprocesses transcribed or typed text, applies named entity recognition for property feature identification, and outputs a structured property feature JSON. The Computer Vision Agent receives the uploaded property property image, applies preprocessing, and runs the trained Regression Model classification model to identify house price patterns with confidence scores. The House Price Prediction Agent receives the combined property feature JSON and property image classification output, queries the Machine Learning module for relevant property context, and generates house price predictions with real estate suggestions and precautionary recommendations. E. Backend API and Frontend The backend handles system logic, AI model communication, and house price prediction processing, implemented in Python using Flask or FastAPI. Backend APIs receive requests from the Gradio frontend, process real estate data, and communicate with AI modules. The Gradio frontend uses a modular layout with data recording interface, property image upload panel, speech-to-text display, agent response panel, and data response playback. Average token consumption per recommendation pipeline execution is approximately 1,800 tokens with response time under 2 seconds per agent call. V. SYSTEM DESIGN A. Use Case Diagram The Use Case Diagram illustrates interaction between users and the Realestate Houseprice Prediction platform. Both General Users and Elderly Users can record data property features, enter text property features, upload property property images, view house price predictions, receive real estate recommendations, and play data responses. The Administrator role has additional privileges for managing property datasets, retraining AI models, and system configuration. The system connects to an external Property Knowledge Base for Machine Learning-based information retrieval during recommendation generation. B. ER Diagram The ER Diagram represents data entities and relationships. The User entity (user_id, name, age) initiates sessions in the Interaction entity (interaction_id, user_id FK, timestamp, input_type). Each interaction includes a VoiceInput record (data_id, audio_data, transcribed_text) and/or PropertyImage record (property image_id, property image_path). Property Feature records are extracted from data and text input. The House PricePrediction entity (prediction_id, house price_name, confidence_score) links to Recommendation records (recommendation_id, precautions, price estimation_suggestion). The PropertyKnowledge entity (knowledge_id, content, embedding) is referenced during Machine Learning retrieval. C. Sequence Diagram The Sequence Diagram illustrates step-by-step interaction during a user session. The user provides data or text input and optionally uploads a property property image via the Gradio UI. The Voice Processing Module converts audio to text and forwards it to the Data Analysis Engine. The Data Analysis Engine extracts property features and passes them to the House Price Prediction Agent. If an property image is uploaded, the Computer Vision Module processes it concurrently. The Prediction Agent queries the Machine Learning Module, which retrieves relevant records from ChromaDB. Results are displayed in the Gradio UI and converted to data output via textto-speech. D. System Architecture Diagram The system architecture follows a layered pipeline structure. The browser-based Gradio UI captures data, text, and property image input. The Python Backend invokes the Voice Processing Module and Data Analysis Engine for text-based input, while the Computer Vision Module handles property image analysis concurrently. Processed outputs feed into the Multi-Agent Orchestrator managing sequential execution of the House Price Prediction and Recommendation Agents. The Machine Learning Module connects to ChromaDB for property knowledge retrieval. Responses flow back through the pipeline to produce price prediction output and data responses via text-to-speech.

PARAGRAPH B¶23-47

TTS Module Converts response to data gTTS, pyttsx3 Dashboard Interactive user interface Gradio, HTML/CSS Table II: Key Functional Modules E. Advantages of Proposed System The Realestate Houseprice Prediction platform eliminates the need for physical property market visits for basic preliminary property feature checking through full multimodal pipeline automation. It provides data-grounded real estate suggestions via Machine Learning, reducing hallucination and increasing property response reliability. The modular multi-agent design ensures quality at each analytical stage. The platform is accessible to elderly and non-technical users through data interaction. The system is deployable on standard hardware without heavy GPU infrastructure requirements. IV. METHODOLOGY A. Overall Workflow The system operates through structured steps. First, the user provides input through data recording, text entry, or property property image upload via the Gradio interface. Second, Speech Recognition converts audio to text. Third, the Data Analysis Engine preprocesses text, extracts property features, and identifies real estate intent. Fourth, Computer Vision analyses uploaded property property images using Regression Model models for house price pattern detection. Fifth, the House Price Prediction Agent synthesizes property feature and property image analysis data and generates house price predictions. Sixth, the Machine Learning module retrieves relevant property knowledge from ChromaDB. Seventh, the Recommendation Agent generates real estate suggestions and precautions. Results are displayed in the chat panel and delivered as data response via text-tospeech. B. Data Collection and Handling The system utilises real estate datasets for training and testing Machine Learning models. House Price property features datasets provide structured property feature-house price mappings for classification model training. Property property image datasets include labelled skin house price property images, wound classifications, and visible infection patterns used to train Regression Model models. Real Estate knowledge documents are preprocessed into text chunks, embedded using sentence-transformer models, and stored in ChromaDB for Machine Learning retrieval. The data handling process includes input validation, missing value handling, data normalisation, property image resizing and enhancement, and serialization of records for embedding and model training. C. Machine Learning Implementation The Machine Learning system forms the property knowledge retrieval backbone. During indexing, property knowledge documents are split into text chunks, embedded using the all-MiniLM-L6-v2 sentence-transformer model, and stored in ChromaDB with source metadata. During retrieval, the user query and extracted property features are embedded, cosine similarity search is performed against ChromaDB collections, and the top-k most semantically relevant property records are returned as structured context for the Recommendation Agent prompt. Score Range Similarity Action > 0.85 High relevance Primary context 0.70 – 0.85 Medium relevance Secondary context < 0.70 Low relevance Exclude from prompt Table III: Machine Learning Retrieval Scoring Criteria D. Multi-Agent Pipeline The multi-agent pipeline consists of four specialised agents operating sequentially. The Voice Processing Agent receives audio input, applies noise reduction, and outputs transcribed text. The Data Analysis Agent preprocesses transcribed or typed text, applies named entity recognition for property feature identification, and outputs a structured property feature JSON. The Computer Vision Agent receives the uploaded property property image, applies preprocessing, and runs the trained Regression Model classification model to identify house price patterns with confidence scores. The House Price Prediction Agent receives the combined property feature JSON and property image classification output, queries the Machine Learning module for relevant property context, and generates house price predictions with real estate suggestions and precautionary recommendations. E. Backend API and Frontend The backend handles system logic, AI model communication, and house price prediction processing, implemented in Python using Flask or FastAPI. Backend APIs receive requests from the Gradio frontend, process real estate data, and communicate with AI modules. The Gradio frontend uses a modular layout with data recording interface, property image upload panel, speech-to-text display, agent response panel, and data response playback. Average token consumption per recommendation pipeline execution is approximately 1,800 tokens with response time under 2 seconds per agent call. V. SYSTEM DESIGN A. Use Case Diagram The Use Case Diagram illustrates interaction between users and the Realestate Houseprice Prediction platform. Both General Users and Elderly Users can record data property features, enter text property features, upload property property images, view house price predictions, receive real estate recommendations, and play data responses. The Administrator role has additional privileges for managing property datasets, retraining AI models, and system configuration. The system connects to an external Property Knowledge Base for Machine Learning-based information retrieval during recommendation generation. B. ER Diagram The ER Diagram represents data entities and relationships. The User entity (user_id, name, age) initiates sessions in the Interaction entity (interaction_id, user_id FK, timestamp, input_type). Each interaction includes a VoiceInput record (data_id, audio_data, transcribed_text) and/or PropertyImage record (property image_id, property image_path). Property Feature records are extracted from data and text input. The House PricePrediction entity (prediction_id, house price_name, confidence_score) links to Recommendation records (recommendation_id, precautions, price estimation_suggestion). The PropertyKnowledge entity (knowledge_id, content, embedding) is referenced during Machine Learning retrieval. C. Sequence Diagram The Sequence Diagram illustrates step-by-step interaction during a user session. The user provides data or text input and optionally uploads a property property image via the Gradio UI. The Voice Processing Module converts audio to text and forwards it to the Data Analysis Engine. The Data Analysis Engine extracts property features and passes them to the House Price Prediction Agent. If an property image is uploaded, the Computer Vision Module processes it concurrently. The Prediction Agent queries the Machine Learning Module, which retrieves relevant records from ChromaDB. Results are displayed in the Gradio UI and converted to data output via textto-speech. D. System Architecture Diagram The system architecture follows a layered pipeline structure. The browser-based Gradio UI captures data, text, and property image input. The Python Backend invokes the Voice Processing Module and Data Analysis Engine for text-based input, while the Computer Vision Module handles property image analysis concurrently. Processed outputs feed into the Multi-Agent Orchestrator managing sequential execution of the House Price Prediction and Recommendation Agents. The Machine Learning Module connects to ChromaDB for property knowledge retrieval. Responses flow back through the pipeline to produce price prediction output and data responses via text-to-speech. VI. IMPLEMENTATION A. Voice Processing Implementation The data processing module is implemented using Python's SpeechRecognition library and OpenAI Whisper for high-accuracy transcription. It reads audio recorded via the Gradio microphone interface, applies noise reduction using the noisereduce library, and transcribes speech to text for downstream Data Analysis processing. Whisper is used for improved accuracy in noisy environments. The pipeline runs in under 1 second for audio clips up to 10 seconds in length. B. Data Analysis Implementation The Data Analysis module is implemented using spaCy and NLTK. It receives transcribed or typed text, applies tokenisation, stopword removal, and lemmatisation. Named Entity Recognition with a custom real estate entity model identifies property features, body parts, and property terms. Intent classification determines whether the user is describing property features, asking for information, or requesting property image analysis. Extracted property feature entities are serialised as a structured JSON for the House Price Prediction Agent. Processing time is under 200 milliseconds per query. C. Computer Vision Implementation The Computer Vision module is implemented using OpenCV for property image preprocessing and a Regression Model model trained using TensorFlow/Keras for property image classification. Input property images are resized to 224×224 pixels, normalised, and augmented during training. The Regression Model architecture follows a ResNet-inspired design trained on dermatological and wound classification datasets. The module outputs a house price class label with a confidence score. Image preprocessing and classification execute in under 1.5 seconds per property image. D. Multi-Agent Pipeline Implementation Agent Input Output Voice Processing Audio recording Transcribed text Data Analysis Agent Text / transcription Property Feature JSON Each agent executes independently and communicates results via a shared state dictionary managed by the backend orchestrator. Error handling includes fallback responses for API failures and confidence thresholding to avoid low-quality predictions being presented to users. E. Challenges and Solutions Challenge Solution Background noise in data input noisereduce preprocessing + Whisper ASR Low property image quality affecting Regression Model Large property image Batch embedding with embedding time progress indicator Table V: Implementation Challenges and Solutions VII. RESULTS AND ANALYSIS A. Functional Results Module Test Case Status Voice Processing Speech-to-text conversion Pass Data Analysis Engine Property Feature extraction from text Pass B. Machine Learning Retrieval Accuracy Query Score Accuracy Fever, cough, fatigue 0.91 High Skin rash with redness 0.87 High Chest pain & breathlessness 0.83 High Joint pain in elderly 0.76 Med-High Table VII: Machine Learning Retrieval Accuracy C. Agent Pipeline Performance Agent Avg. Time Quality Voice Processing < 1.0 sec High transcription accuracy

PARAGRAPH B¶44-62

embedding time progress indicator Table V: Implementation Challenges and Solutions VII. RESULTS AND ANALYSIS A. Functional Results Module Test Case Status Voice Processing Speech-to-text conversion Pass Data Analysis Engine Property Feature extraction from text Pass B. Machine Learning Retrieval Accuracy Query Score Accuracy Fever, cough, fatigue 0.91 High Skin rash with redness 0.87 High Chest pain & breathlessness 0.83 High Joint pain in elderly 0.76 Med-High Table VII: Machine Learning Retrieval Accuracy C. Agent Pipeline Performance Agent Avg. Time Quality Voice Processing < 1.0 sec High transcription accuracy Data Analysis Agent < 0.2 sec Accurate property feature extraction Computer Vision < 1.5 sec Reliable property image classification Prediction Agent ~1.2 sec Actionable health guidance Full Pipeline ~4.0 sec Real Estate-ready output Table VIII: Agent Pipeline Performance D. System Performance Parameter Observation End-to-End Pipeline Time ~4.0 seconds average ChromaDB Query Time < 400 ms Image Classification Time < 1.5 seconds Voice Transcription Time < 1.0 second Dashboard Load Time < 3 seconds TTS Response Generation < 0.8 seconds Table IX: System Performance E. User Experience Analysis The Gradio dashboard provided a smooth and intuitive user experience across all test scenarios. The data recording interface was accessible and responsive for both technical and non-technical users including elderly participants. The property image upload functionality performed reliably across different property image formats and sizes. The text-to-speech data response added significant accessibility value. The dashboard load time of under 3 seconds and sub-4-second end-to-end pipeline execution met performance expectations for a real-time real estate assistant platform. VIII. DISCUSSIONS A. Advantages of the Proposed System The Realestate Houseprice Prediction platform offers several key advantages over traditional real estate applications. It eliminates the need for physical visits for basic property feature checking — once a user records data property features or uploads an property image, the entire pipeline executes automatically without further intervention. By grounding all real estate recommendations in semantically retrieved data from ChromaDB, the system ensures suggestions are based on verified property knowledge rather than generic AIgenerated content, significantly reducing hallucination and increasing buyer trust [9]. The multi-agent architecture ensures each cognitive stage is handled by an agent specifically designed for that task, producing higherquality outputs than a single-agent approach. The Gradio data interface allows elderly buyers without technical expertise to interact with complex AI real estate analysis through natural data communication. The system can run on a standard laptop without requiring GPU infrastructure or complex cloud deployment. B. Limitations of the System Several limitations were identified during development and testing. The system provides preliminary guidance only and is not intended to replace professional clinical price prediction. Prediction accuracy depends on the quality and diversity of training datasets. Voice recognition accuracy decreases in high-noise environments despite preprocessing. Image classification performance depends on property image quality and lighting conditions [11]. No user authentication No multi-user access control implemented Table X: System Limitations C. Real-World Applicability Rural and remote users can access the platform for basic real estate guidance without travelling to property markets. Elderly users can interact naturally through data, reducing barriers posed by complex digital interfaces. Real Estate workers can use the platform as a preliminary triage tool before formal consultation. The platform also serves as a practical demonstration of multimodal AI integration for academic research and educational settings in AI and real estate engineering. IX. FUTURE SCOPE The Realestate Houseprice Prediction platform has been designed as a scalable and modular system allowing for continuous enhancement. Several significant improvements are planned for future versions. • Real-Time Data Integration: Connecting to property market EHR systems, pharmacy APIs, and wearable health devices to enable live health data analysis and personalised continuous real estate recommendations. • Advanced House Price Prediction Models: Integrating timeseries health data analysis and training models on larger, more diverse property datasets to improve prediction accuracy and house price coverage. • Telemedicine and Video Consultation: Adding real-time video consultation for agent-buyer communication and enabling online appointment scheduling and emergency consultation support. • Multi-Language Support: Implementing data recognition and text-to-speech in regional and international languages to expand accessibility for diverse global user populations. • Enhanced Property Image Analysis: Extending Computer Vision capabilities to support X-ray, MRI scan, CT scan, and blood report analysis using advanced Deep Learning architectures. • Mobile Application Development: Developing Android and iOS mobile applications with real-time health alerts, data interaction, and report download capabilities. • Cloud Deployment and Scalability: Docker containerisation, Kubernetes orchestration, and scalable vector databases such as Pinecone or Weaviate for large-scale property knowledge bases. • AI-Based Mental Health Support: Detecting stress, anxiety, and depression property features through data pattern analysis and conversational interaction, providing mental wellness guidance. X. CONCLUSION The AI-Powered Multimodal Real Estate Assistant (Realestate Houseprice Prediction) presents an effective and scalable solution for automating preliminary real estate guidance and improving real estate accessibility in modern digital environments. The system successfully integrates Retrieval-Augmented Generation with a multi-agent AI architecture to deliver data-grounded, multimodal real estate assistance through an interactive data and property image-capable dashboard interface. The project demonstrates the successful implementation of a full-stack AI pipeline using Speech Recognition for data processing, spaCy and NLTK for Data Analysis-based property feature analysis, OpenCV and Regression Model models for property property image classification, ChromaDB for semantic property knowledge retrieval, and LLM integration for recommendation generation. The Gradio frontend provides an accessible and functional interface for both technical and non-technical users, including elderly buyers, with natural data interaction, property property image upload, AIpowered house price prediction, and text-to-speech data response output. Results from system testing validate the platform's ability to deliver real estate guidance with an end-to-end pipeline execution time of approximately 4 seconds and consistently high Machine Learning retrieval relevance scores. The system acts as a smart preliminary real estate assistant capable of providing valuable guidance anytime and anywhere, representing an innovative step toward intelligent real estate solutions and highlighting the growing importance of AI-powered multimodal digital real estate systems in bridging the real estate accessibility gap.