Data Engine
A self-updating data operating system — it turns raw company data into reliable, versioned, traceable SQL knowledge with minimal manual work.
What it does
Ingest any company data (starting with spreadsheets), convert it into structured SQL tables, keep rich metadata about what each dataset means, and maintain a knowledge layer about how datasets relate to the business.
How it works
- Upload — raw files are kept faithfully in object storage.
- Propose — an action agent (Claude) reconciles the upload against the current knowledge base.
- Check — a checker agent (Gemini) reviews and feeds back, until satisfied.
- Approve — a human reviews the proposal and approves, requests changes, or rejects.
- Apply — a deterministic executor commits the change and logs everything.
Core principles
- Idempotent, deduplicated ingestion
- Alias & entity resolution
- Structured action plans, never freeform SQL
- Snapshots, rollback & full auditability
- Least-privilege execution
- Human-supervised now → progressive autonomy later