The OMEGA framework generates executable ML classifiers that outperform scikit-learn baselines across 20 benchmark datasets. It integrates structured meta-prompting with code generation to automate the research cycle from idea to implementation. This system enables practitioners to discover novel algorithms via the omega-models Python package. It reduces manual iteration in classifier design.