EdotEnv Uses Quant Trading Workflows as Self-Improving RL Benchmarks for LLMs

Startup EdotEnv, founded by former quant traders Rui and Michael and backed by Y Combinator (S26), is building reinforcement learning environments based on professional quantitative trading workflows to train and evaluate large language models. The core idea is that financial markets grow more efficient over time as inefficiencies are exploited, making them a naturally self-scaling benchmark that avoids the saturation problem common in static AI evals. Their environments task LLM agents with real-world quant research steps — including feature engineering, portfolio design, backtesting, and strategy adaptation — using actual market data and verifiable rewards. Early testing of state-of-the-art models revealed notable weaknesses: agents tend toward broad, shallow research rather than deep iteration, and lack basic trading intuition. EdotEnv plans to sell these continuously evolving environments to AI labs, researchers, and enterprises focused on improving agents' long-horizon planning and continual learning capabilities.