ML / Trading

Portfolio

Portfolio is a classic machine-learning project centered on XGBoost ranking models for crypto strategy research. I built it as a crypto research and paper-trading platform that combines automated market data scraping, machine-learning ranking models, and live experiment tracking. The system collects historical market data, trains and evaluates multiple strategy variants, and powers an interactive dashboard to compare model performance and portfolio decisions across venues in real time.

ActivepublicUpdated May 26, 2026

Visit project Back to projects

Screenshots

Portfolio crypto ranker landing page with live paper trading performance — The public landing page for the crypto ranker, showing live paper-trading performance, venue context, and a plain-English model explanation.

Portfolio cockpit strategy table comparing model variants — The live cockpit table for comparing XGBoost strategy variants, venues, portfolio decisions, P&L, alpha, evidence coverage, and holdings.

Problem

Crypto strategy research is fragmented across data collection, model evaluation, and execution testing, which makes it hard to compare strategies consistently.

Solution

A unified platform that scrapes and curates market data, trains and compares multiple ML strategy variants, and tracks paper-trading outcomes in a single interactive environment.

Architecture

An end-to-end ML research stack with automated market data ingestion, XGBoost ranking-model training and evaluation, paper-trading simulation, and a live dashboard for cross-venue strategy comparison.

Lessons learned

Classic ML still delivers strong value when data quality, evaluation discipline, and experiment tracking are done consistently.