In a business intelligence (BI) project, it’s not the dashboard that costs the most, but everything that comes before it. Connecting data sources, understanding tables, identifying keys and relationships, and then building the data model: this work takes weeks and requires the involvement of experts.

SEIKAI, in collaboration with LIS (CNRS) and AMSE, is conducting an R&D program to automate this process. Deterministic algorithms (key detection, Levenshtein distance, entropy, etc.) calculate and verify the model, and LLMs semantically enrich it to adapt it to the specified analysis requirements.

In 10 Minutes: Our initial findings on ERP and CRM data, the remaining challenges (noisy data, business semantics, explainability), and the outlook—toward traceable dbt models and an open Bronze-to-Gold standard.