Leeroo open-sources Kapso: enterprise data work becomes an experiment loop with a git-shared lesson bank
Synopsis
Leeroo has open-sourced Kapso, an engine for enterprise data work: given an objective and success criteria, it proposes hypotheses, runs experiments in a sandbox, and writes the results back into a lesson bank that carries credence scores and applicability conditions. The bank is built on git, so teams can share and update it, and the system is model-agnostic and coding-agent-agnostic, running a shorter experiment loop on the target stack to adapt the best solution at deploy time.
Interpretation
Enterprise data work is organized as a repeatable experiment loop: the system proposes hypotheses around an objective, runs experiments in a sandbox, and accumulates know-how from every run. The prior path was consultants gathering context in meetings, or teams defining a business ontology upfront and fitting data to it, which the source says has no chance of execution in quarters and sometimes years. One described enterprise project moved a company off its old data estate onto a lakehouse with more than 30 source-system families, thousands of tables and transformations, and hundreds of BI reports, and no individual or AI agent knew beforehand how that estate was connected and operated.
Before each new experiment, an ideation module proposes candidate ideas informed by earlier lessons, re-ranks them by expected return against the budget each would use, and sends the top-ranked ones to the implementor. Each lesson in the bank carries a credence score, conditions on where it applies, and the evidence behind it, so proposed ideas fit the environment better than generic hypotheses from an LLM or retrieval-augmented generation. A verifier module that no experiment can access, and therefore cannot cheat, checks each result, and after every run learn() updates the bank's scores and boundaries using that run's evidence; changes to the bank can be A/B tested by running otherwise identical builds with the previous and updated banks.
Perspective
The demonstrated scope is the iterative experimentation and deployment workflow for enterprise data environments, including a migration project spanning more than 30 source-system families, thousands of tables and hundreds of BI reports, plus deployment adaptation to target stacks such as Docker, Modal, BentoCloud and LangGraph; the intended reader is an engineer building enterprise data systems and pipelines.
The source is mainly a project narrative and system description; it reports no experiment counts, success rates or baseline comparisons, and it does not conclude whether the lesson bank remains applicable when transferred across environments.