Day Trading Forecasting System
A 6-stage machine learning pipeline that ingests market data from multiple providers, engineers 21 technical indicators, labels future price movements across 4 time horizons, trains per-horizon classifiers, serves sub-200ms predictions, and executes trades through a multi-broker trading client — all with complete audit trails.
This system answers one question for 15 actively-monitored US equities: “Will this stock go UP, DOWN, or stay FLAT in the next 5 minutes, 25 minutes, 75 minutes, or 5 hours?”
Every 10 minutes during market hours, an automated pipeline fetches real-time price data from 4 providers, computes 21 technical indicators across 5 categories (price-based, momentum, volatility, trend, and derived), classifies future price movements across 4 time horizons, trains per-horizon RandomForest classifiers, generates predictions in under 200ms, and executes trades through multi-horizon consensus — requiring agreement across multiple timeframes before placing any order.
The system currently monitors 15 symbols spanning mega-caps, growth stocks, financials, and small-caps. It runs on a cloud stack of PostgreSQL, Azure Blob Storage, FastAPI, N8N, and Docker at an estimated cost of $20–40/month using free tiers. Three broker integrations — Alpaca (production), Interactive Brokers (testing), and Kotak Securities (in progress) — provide multi-market capability through a provider-agnostic interface.
Explore the detailed documentation below to understand how each component works, how the paper trading implementation operates, and how the ML approach compares to traditional trading strategies.