The Two Unsolved Problems in Quant Research
If you've spent any time backtesting trading strategies, you've probably run into both of these:
Problem 1: Overfitting is embarrassingly easy. Most backtesting tools will happily show you a 40% CAGR strategy that falls apart the moment it touches unseen data. The backtest looked great because you — consciously or not — optimised in-sample and called it done. Walk-forward validation exists to catch this, but it's tedious to wire up manually, so most people skip it.
Problem 2: Existing quant tools are impossible for an AI agent to drive. Web-UI backtesting platforms have no CLI surface. Raw Python frameworks are powerful but their APIs are wide and stateful — asking Claude Code to "explore strategies overnight" means the agent would have to parse Python tracebacks, infer what broke, and mutate code files in a loop. That's fragile. It also means you need to babysit it.
I built AlphaForge to solve both at once.
