Show HN: Kapso – Iterative learning from enterprise data

Kapso is an open-source system for solving enterprise data and engineering problems through repeated experiments rather than relying on a fixed business ontology or documentation search. Its creators argue that large, tangled data environments can only be understood by investigating code, models, documentation, and data, testing hypotheses, and letting each discovery guide the next step. They describe a migration involving more than 30 source-system families, thousands of tables and transformations, and hundreds of reports, where no person or AI agent initially understood how the estate worked. The same iterative approach applies to building and optimizing software: what succeeds depends on the particular environment.

Kapso maintains a bank of lessons, each with evidence, applicable conditions, and a confidence score. For each objective, it proposes and ranks experiments by expected value and cost; an isolated verifier checks results, and successful or failed runs update the lessons. Teams can test lesson-bank changes, share banks through Git, create domain-specific harnesses, and deploy solutions adapted to target platforms. The system can also propose changes to evaluation criteria, subject to verifier approval. Its creators report first place on RelBench, a top result on OpenAI’s MLE-bench, and strong performance in AI competitions, including a top-three AI-system finish at IOAI 2026. They also say Kapso helped complete a decades-old enterprise data-estate migration in weeks after the project had stalled for over a year, while revealing opportunities to reduce costs.