Realestica - AI Powered Platform for the Jakarta Property Market

01 / Process

How the work unfolded

To address the challenges in the property market, Realestica was developed as a comprehensive platform that integrates data and machine learning to provide transparency and efficiency.

Process & Research

The project began with an analysis of market challenges and existing competitors like Rumah123 and Pinhome. Their approaches were deemed too general and not fully effective as they do not adequately consider detailed house features or Points of Interest (POI). Based on various academic studies, we identified that crucial factors—such as location, physical specifications (building and land area), land certificate type, housing facilities, and proximity to public amenities—have a significant impact on property values but are often underutilized by current platforms.

Architecture & Implementation

The Realestica system is built on a solid architecture. The process begins with a Web Scraper that collects public property data. This raw data then undergoes a Preprocessing & Cleaning stage before being stored in a database. This clean data is subsequently used to train our Machine Learning Model. The business logic is handled by a Backend built with FastAPI, which serves three core components:

1. A Prediction Service for estimating property prices.
2. Recommender System for personalized suggestions.
3. Property Management API to handle property data.The results are then presented to the user through the Realestica Frontend.

Results & Validation

The price prediction model demonstrated strong and relevant performance, achieving an R² score of 0.72and a Mean Absolute Percentage Error (MAPE) of 16.27%. These results indicate that the model has powerful predictive capabilities and a low error rate, making it a reliable tool for supporting decision-making in the property market.