Technology

Weco AI

Weco AI automates the machine learning exploration loop: it proposes, implements, and empirically tests hundreds of code variations overnight.

Weco AI fields an LLM-driven agent, AIDE, to automate the rigorous trial-and-error inherent in ML engineering. This system employs a 'Solution Space Tree Search' methodology: it proposes code edits, executes them locally, and refines solutions based on continuous evaluation metrics. The platform is proven: its predecessor, AIDE ML, secured 4x more medals than the next best autonomous agent on OpenAI’s MLE-Bench across 75 Kaggle competitions, demonstrating superior performance in code optimization and model training. Weco handles GPU kernel tuning, architecture search, and prompt design, allowing ML teams to convert compute power directly into measurable wins, faster and with less manual effort.