Pipeline design
The project is organized as explicit stages so each artifact can be inspected, reproduced, and reused downstream.
Portfolio Demo
Market Analytics Pipeline
A reproducible analytics project that turns public commodity time series into engineered features, baseline forecasts, comparison plots, and grounded local-LLM market notes.
Snakemake
SQLite
Ridge forecasting
Ollama report
Why This Project
One demo, several layers of pipeline thinking
Cross-asset comparison
Coffee, cocoa, tea, and sugar are processed with the same flow so the demo can compare behavior, model quality, and reporting outputs side by side.
Grounded LLM usage
Ollama is used only at the reporting layer, with structured metrics and recent forecast facts as the source of truth.
Pipeline Stages
From raw series to analyst note
Ingest + Features
Real samples from the pipeline
Raw preview
Feature preview
Training
Cross-commodity model summary
Comparison table
Architecture
How the pipeline is stitched together
1
FRED fetch
Monthly commodity series downloaded into raw CSVs.
→
2
SQLite store
Each asset is persisted locally for downstream processing.
→
3
Feature engineering
Lagged returns and rolling volatility features are generated.
→
4
Model + plots
Ridge baselines, metrics, predictions, and comparison visuals.
→
5
Ollama note
Grounded markdown reports generated from structured facts.
Figures
Visual diagnostics built by the pipeline
Local LLM Reports