Show HN: Kapso – Iterative learning from enterprise data (github.com)

Kapso, an innovative open-source engine developed by Leeroo, has made waves in the AI/ML community by achieving significant milestones in recent benchmarking competitions. Not only did Kapso win the Grand Master Trophy at the International Olympiad in AI (IOAI²), outperforming all human and AI competitors with a score of 536.07, but it also dominated the RelBench leaderboards in classification, regression, and recommendation tasks, highlighting its robust capabilities in handling relational data. Furthermore, Kapso topped both the MLE-Bench and ALE-Bench benchmarks, showcasing its versatility in machine learning competitions and long-horizon algorithmic discovery problems. What sets Kapso apart is its iterative learning approach, which allows it to conduct smart experimentation, deploy optimal solutions, and continuously learn from each run. This engine utilizes a tree of experiments to propose and verify hypotheses while building a knowledge bank of lessons derived from experimental outcomes, complete with credibility scores for each lesson. Its model-agnostic nature ensures compatibility with various coding agents, making it a versatile tool for data scientists and ML engineers aiming to refine their implementations efficiently and effectively.