Show HN: Norma – build good datasets (using an objective) (norma.grouplabs.ca)

Norma is a data-prep layer that sits between raw warehouse tables and ML/BI tools, automatically discovering relevant tables, building consolidated, leakage-safe feature representations, and outputting model- and analytics-ready datasets. Instead of manual join paths and hand-tuned feature engineering, Norma explores your schema, constructs features across sensors/events/ETL streams, validates datasets to guarantee zero leakage, and exports data or APIs for downstream tools like AutoGluon, XGBoost, TabPFN and PowerBI. It supports an integrated SQL+Python workflow, a natural-language suggestion interface, fast DuckDB execution for millions of rows, and direct Databricks integration for existing infra. For practitioners this reduces iteration time and risk: automatic feature synthesis, representation quality scoring, multi-bandit and 5‑fold validation, drift detection, lineage visualization and automated leakage prevention tighten the feedback loop and improve model reliability. Planned and shipping features—AutoML integration, differential-privacy LLM transforms, synthetic data generation, shared datasets and federated optimization—signal a move toward self-optimizing pipelines that maintain privacy and governance while boosting model performance. Norma’s combination of schema-aware discovery, validation-first engineering, and tight infra integrations makes it a practical tool for teams struggling with messy warehouses and data-leakage pitfalls.