Core Principle

No AI models training on company data.

We embed controlled access via Azure OpenAI with maxium privacy enabled:

  • Analyze metadata, schemas, queries
  • Infer lineage
  • Document the current state

This significantly accelerates the assessment process by eliminating manual effort securely.

Why This Matters

  • Assessment is a reasoning and documentation problem, not a prediction problem
  • Value comes from orchestration (prompt-engeneering), governance (enough context), and human validation (Atlan tags)
  • Training is unnecessary for assessment while expensive and slow

Strategic Advantage

  • Fast time-to-value (weeks, not months). Can allocate more human effort on bottlenecks in processes.
  • Privacy-peserving - AI scans config files only, no data files -

We start with inference. We keep fine-tuning as an option, not a dependency.