Computer Science + Data Science & AI

Inflation forecasting with context, foundation models and MCP.

A double Final Degree Project exploring when economic news, institutional communication and external context improve inflation forecasts, and turning that research workflow into a reproducible web platform.

Summary slide showing that there is not one winner across inflation forecasting tasks.

Core takeaway

More context is not automatically better.

Context helps when it is aligned with the target series, available at the right time, and integrated without leakage or noise. Simpler baselines still matter.

Forecasting evidence

Results across Spain CPI, Global CPI and European HICP.

Strongest contextual result for Global CPI, showing Chronos-2 with contextual information beating the baseline.
Spain CPI slide showing that classical ARIMA baselines still win at longer horizons.
European HICP slide showing a strong foundation-model base with a small contextual gain.

Software platform

From experiment files to an inspectable forecasting system.

Backend Frontend PostgreSQL MongoDB Docker MLflow MCP Drift analysis
Application screenshot showing the model comparison dashboard.
Application screenshot showing the what-if simulator.
Application screenshot showing inflation news and contextual signals.

Final documents

Two theses, one integrated project.

Data Science & AI

Forecasting inflation with contextual signals

Model comparison across statistical, deep learning and foundation time-series models, with MCP-driven contextual inputs.

Open PDF

Computer Science

Reproducible forecasting experimentation platform

Full-stack web platform with forecasting adapters, experiment tracking, databases, Docker deployment and MCP integration.

Open PDF