Back to projects
In productionData EngineeringTeam · Data Engineer & ML Developer
HOMEPEDIA
Data-driven platform combining Machine Learning and French open data (DVF, demographics, crime, education, housing) to estimate price per m² by department, property type, season and land area. FastAPI REST API with 10 thematic endpoints. Role: data architecture, ETL pipelines, regression model training and API implementation.
10
Thematic API endpoints
5+
Data sources integrated
2
ML regression models
Architecture
- Open data
- DVF
- Demographics
- Crime
- Education
- Housing
- ETL pipelines
- PySpark
- Pandas
- NumPy
- Storage
- PostgreSQL
- SQLAlchemy
- Models
- Scikit-learn: 2 regression models
- Joblib
- Serving
- FastAPI: 10 endpoints
- Pydantic
- Docker
Stack
PythonFastAPIPostgreSQLSQLAlchemyScikit-learnPySparkPandasNumPyPydanticJoblibDocker
- Heterogeneity of French public data sources requiring complex normalisation
- Feature engineering tailored to real-estate market specifics (seasonality, property types)
- Matching geographic data across different reference systems (departments, municipalities)
- Maintaining consistent one-hot encoding between training and production inference
- Modular ETL pipelines with PySpark for data cleaning and transformation
- Categorical variable encoding with saved model columns to ensure consistency
- Decoupled API architecture enabling independent evolution of endpoints and models
- Heuristic price adjustments based on property condition (existing building or bare land)
- API in production delivering real-time real-estate price estimates
- Consolidation of French socio-economic data into a single, structured database
- Serialised ML models deployed via Docker
A question about this project?
Contact me