Financify — Inflation Forecasting
A Bangkit capstone project: an LSTM model forecasting inflation across Indonesian cities and periods, served through a Flask application layer that the team's mobile app consumes.
- Role
- Machine Learning Engineer
- Year
- 2023
- Fields
- AI / ML / Data / Optimization
- Stack
- Python
- TensorFlow / Keras
- LSTM
- Pandas
- Flask
- Docker
What shipped
LSTM forecasting
Sequence modeling tuned to inflation's temporal structure.
Multi-city coverage
Forecasts across Indonesian cities and periods, not one aggregate curve.
Flask application layer
Predictions served through an API the rest of the team builds on.
Docker packaging
The service ships in a container, ready for deployment.
What changed
- Forecasts inflation across Indonesian cities and periods.
- Built within the Bangkit Machine Learning learning-path team.
- The Flask layer lets the team's mobile application consume the model.
Process details
Forecasting an economy, city by city
Inflation moves differently in every Indonesian city and every period. A single global curve would be useless — the model had to learn temporal patterns per city and serve them through an interface other teams could build on.
As the machine-learning engineer, my work sat between data and product: prepare the series, design an LSTM that respects their sequence structure, and expose forecasts through a Flask layer the mobile team could integrate.
Series in, forecasts out, API in between
Data
Indonesian inflation series, prepared
City-level inflation data cleaned and shaped with Pandas into sequences the model can learn from.
Model
An LSTM for temporal patterns
A TensorFlow/Keras LSTM learns the sequence structure of inflation across cities and forecast periods.
Delivery
Flask and Docker as the handoff
A Flask application layer exposes predictions, packaged with Docker so the team's mobile app integrates against a stable interface.