Skip to content

README.md

Latest commit

 

History

History
92 lines (71 loc) · 3.06 KB

README.md

File metadata and controls

92 lines (71 loc) · 3.06 KB

AlphaStream

ML-powered trading signals for futures markets. Real models, real data, no fake metrics.

Architecture

AlphaStream/
├── backend/          # Python FastAPI + ML pipeline
│   ├── alphastream/
│   │   ├── api/      # FastAPI server
│   │   ├── data/     # Market data fetcher (yfinance)
│   │   ├── features/ # Feature engineering (45 features, no leakage)
│   │   ├── models/   # Training + inference
│   │   ├── signals/  # Signal generation
│   │   └── config/   # Settings
│   ├── models/       # Trained model artifacts (.joblib)
│   ├── train.py      # Training CLI
│   ├── schema.sql    # Supabase database schema
│   └── Dockerfile    # Railway deployment
└── frontend/         # Next.js 16 + React 19
    ├── app/          # 33 pages (landing, dashboard, auth, settings)
    ├── components/   # 122 components
    ├── contexts/     # Auth + realtime (connected to real API)
    └── lib/          # API client, types, utilities

What's Real

  • 7 symbols: ES, NQ, CL, GC, BTC, ETH, RTY
  • 4 model types per symbol: XGBoost, LightGBM, Random Forest, Gradient Boosting
  • 28 trained models with walk-forward validation (5 folds, purge gap)
  • 45 technical features — all backward-looking, zero data leakage
  • 11,400+ hourly bars per symbol (2 years of data)
  • FastAPI server with real endpoints: signals, models, prices, backtest

Quick Start

Backend

cd backend
python -m venv .venv
source .venv/Scripts/activate  # Windows
pip install pandas numpy scikit-learn xgboost lightgbm joblib yfinance fastapi uvicorn pydantic python-dotenv loguru httpx

# Train models
PYTHONPATH=. python train.py

# Start API server
PYTHONPATH=. uvicorn alphastream.api.server:app --port 8000

Frontend

cd frontend
npm install
npm run dev

Visit http://localhost:3000 (frontend) and http://localhost:8000/docs (API docs)

API Endpoints

Method Path Description
GET /health Health check + model status
GET /v1/signals Latest signals (all symbols)
GET /v1/signals/{symbol} Signal for specific symbol
GET /v1/models Model performance metrics
GET /v1/prices/{symbol} Historical price data
POST /v1/backtest Run backtest on historical data

Deployment

  • Frontend: Vercel (auto-deploy from GitHub)
  • Backend: Railway (Docker)
  • Database: Supabase (run schema.sql)

Honest Performance

Models achieve ~50% directional accuracy on hourly futures data with basic technical features. This is realistic — futures markets are highly efficient at this timeframe. Edge comes from:

  • Ensemble consensus (4 models voting)
  • Confidence-weighted signals
  • Risk management (ATR-based stops)

Tech Stack

Backend: Python 3.12, FastAPI, scikit-learn, XGBoost, LightGBM, yfinance Frontend: Next.js 16, React 19, TypeScript, Tailwind 4, Recharts, Framer Motion, Radix UI Infrastructure: Vercel, Railway, Supabase, Stripe