Market Research · Live
Rankoon
Rankoon grew out of my Portfolio project, a classic machine-learning setup with XGBoost ranking models for crypto strategy research. It is now a public site with live signals across time horizons, sector rotation, daily model rankings, paper portfolios, and transparent historical hit rates. Everything runs on public market data and simulated money. It is research, not investment advice.
- My role
- Creator, ML engineer, and product builder
- Started
- Updated
- Visibility
- Public
- Built with
- Python · XGBoost · Walk-forward validation · Automated data collection · Paper trading · Azure
- model rankings across the ranked universe
- Daily
- validation with published hit rates
- Walk-forward
- simulated money, not investment advice
- Paper only
Problem
Crypto research is fragmented across data collection, model evaluation, and execution testing, which makes it hard to compare strategies consistently or to know whether a signal ever worked.
Solution
One public product that collects data, ranks assets with ML models, publishes signals with their historical hit rates, and tracks paper portfolios against the market.
Architecture
Scheduled collectors ingest public market data, XGBoost ranking models score assets daily, a walk-forward evaluation tracks hit rates honestly, and paper portfolios follow a simple simulated policy. A web app surfaces signals, rankings, research pages, and portfolio performance.
Lessons learned
Classic ML still delivers when data quality, evaluation discipline, and honest reporting are consistent. Showing hit rates next to signals builds more trust than any model description.
More screenshots (2)
machine-learning · ranking-models · crypto · market-research