What is alpha-forge?

Alpha Forge is an infrastructure project for building AI-driven systematic trading systems. Systematic trading means using software and statistics to make investment decisions automatically, rather than relying on a person's instincts in the moment. This project connects several components: reading financial language to find trading signals, training machine learning models, managing a portfolio of strategies, and placing orders through a brokerage. The language side works by feeding filings, earnings calls, news articles, and similar text into a set of AI agents. Those agents extract structured signals from the text, a process the project calls "text-event" features. These signals become inputs for prediction models alongside traditional market data like prices and trading volumes. The machine learning layer combines several model types, including gradient boosting and learned numeric representations, to forecast which securities might perform well. Once a strategy is developed, the system guides it through a staged promotion process before it touches real money. It goes through shadow mode, where it runs silently alongside live markets without placing trades, then paper trading with simulated orders, then a soak period, and finally a controlled live deployment. Each stage has checks that must pass before the strategy advances. This structure is intended to prevent an untested or misbehaving strategy from reaching a real brokerage account. For execution, the project integrates with Interactive Brokers. Before any order goes out, the system runs pre-trade checks and can halt everything via kill switches if something looks wrong. It also keeps detailed journals of events for audit purposes and reconciles expected positions against actual brokerage state to catch discrepancies. The portfolio layer handles situations where multiple strategies are running at once. It collects their individual trade proposals and resolves them into a single set of orders for the account, applying risk limits in the process. The project is built with Python 3.11 and includes Docker support. The README is extensive but was truncated in this source.