Leeroo open-sources Kapso, an engine that iteratively learns from enterprise data experiments

Leeroo, a team of former LinkedIn and Meta AI builders, has open-sourced Kapso, an engine that runs experiments, deploys the best-performing solution, and learns from each run. The company argues that enterprise data work is inherently iterative, requiring hypotheses to be tested in specific environments rather than defined upfront or handled through generic RAG. In one enterprise project, the team used a similar investigate-and-experiment approach to map a lakehouse migration spanning more than 30 source-system families and thousands of tables. Kapso maintains a bank of lessons, each with a credence score, applicability conditions, and supporting evidence, which guides ideation for new experiments. A separate verifier module evaluates results so experiments cannot manipulate their own success criteria.