House price prediction
Train a machine learning model capable of estimating house prices from property characteristics.
MACHINE LEARNING / MLOPS
House price prediction, built beyond the notebook.
A production-style machine learning system designed to demonstrate how a model moves from training and validation through automated deployment and API-based inference.
THE PROJECT
From model to system.
The goal is not simply to produce an accurate prediction. The project demonstrates the engineering required to make the model reproducible, testable and deployable.
Structured features
Property data is transformed into model-ready features before being passed through the training pipeline.
Automated workflow
Training, validation, registration and deployment are connected through an automated CI/CD workflow.
API predictions
The trained model is packaged with BentoML and exposed through an inference API used by the prediction interface.
SYSTEM ARCHITECTURE
From source code to inference.
The system separates model development from deployment while keeping both workflows automated and reproducible.
SOURCE
01 / SOURCE CONTROL
GitHub
Application code, model code and CI/CD workflow definitions
↓
CONTINUOUS INTEGRATION
02 / AUTOMATION
GitHub Actions
Triggered by changes to the development branch
→
03 / TRAINING
Model Training
Features are retrieved and a new model candidate is trained
→
04 / VALIDATION
Deepchecks
Automated validation determines whether the candidate can progress
↓
MODEL LIFECYCLE
05 / TRACKING
MLflow
Experiments, metrics and model artifacts
→
06 / REGISTRY
Model Registry
Versioned model selected for downstream deployment
CONTINUOUS DELIVERY
07 / PACKAGING
BentoML
Registered model packaged as an inference service
→
08 / CONTAINER
Docker
Reproducible container image for deployment
→
09 / REGISTRY
AWS ECR
Versioned container images stored in AWS
↓
INFERENCE
10 / SERVING
Prediction API
The deployed BentoML service receives property features
and returns a house-price prediction
MLOPS WORKFLOW
What happens when code changes?
01
Code is pushed to GitHub
Changes to the development branch trigger the continuous integration workflow.
02
The model is trained
The CI workflow executes the training pipeline, producing a new model candidate from the available feature data.
03
The candidate is validated
Automated validation checks are executed before the model is allowed to progress through the pipeline.
04
The model is registered
A validated model is registered so that the deployed artifact can be associated with a specific model version.
05
CD packages the service
The deployment workflow packages the model-serving application and builds its container image.
06
The image is pushed to ECR
The container image is published to Amazon ECR, providing a versioned deployment artifact.
07
The model becomes an API
BentoML exposes the packaged model as an inference service that can receive property data and return predictions.
TECHNOLOGY
Tools used across the lifecycle.
IMPLEMENTATION STATUS
What is actually running?
The portfolio distinguishes implemented components from infrastructure that is still being developed.
COMPONENT
STATUS
Model training
Implemented
Feature store
Implemented
Model validation
Implemented
MLflow experiment tracking
In development
Model registry
Implemented
Continuous integration
Implemented
Continuous delivery
Implemented
BentoML serving
Implemented
Docker containerization
Implemented
AWS ECR
Implemented
Prediction API
Implemented
Production monitoring
Planned
LIVE INTERFACE
Test the model.
Use the prediction interface to send property features to the model-serving API and inspect the returned prediction.