Database Modelling

[isoctcolo]

Hello, I tested Vanna to achieve a data modelling solution based on Data Vault 2.0 modelling. It worked well out of the box. What would be the best solution to enhance data modelling with Vanna?

[mergisi]

Great question! For enhancing data modelling with Vanna, especially with Data Vault 2.0, here are some approaches that work well:

1. Training with DDL + Documentation: Feed Vanna your complete DDL statements along with business context documentation. Data Vault's Hub/Link/Satellite structure benefits from clear relationship documentation.
2. Custom Embeddings: Consider fine-tuning the embedding model with your domain-specific terminology. Data Vault has specific naming conventions (HUB_, LNK_, SAT_) that models can learn.
3. Query Examples: Train with a diverse set of example queries covering different join patterns between Hubs, Links, and Satellites. This helps the model understand the typical query patterns in DV2.0.
4. Schema Context: Use vn.add_documentation() to add context about your business keys, hash keys, and load date patterns.

vn.add_documentation("HUB tables contain business keys and hash keys")
vn.add_documentation("SAT tables store historical attribute changes")
vn.add_documentation("LNK tables represent relationships between hubs")

If you find the setup overhead too much for simpler use cases, tools like ai2sql.io offer a more streamlined approach where you can generate SQL without extensive training - though for complex Data Vault scenarios, a customized solution like yours with Vanna gives more control.