GitLab's conversational AI needs to generate executive briefing docs and account plans on demand — requiring more than just structured data. Amie shares their three-part architecture: structured data from warehouses (the foundation they've had for years), vectorized unstructured data that marries seamlessly with structured sources, and context pulled from semantic layers, dbt docs, metadata, and business glossaries.
Key takeaways:
Conversational AI for enterprise use cases requires three components working together, not just one data source
Structured data from warehouses provides the foundation — the single source of truth companies have built for years
Unstructured data must be vectorized to combine seamlessly with structured sources for comprehensive AI responses
Context comes from multiple places: semantic layers (dbt docs), metadata repositories, and business glossary definitions
Account IQ and Customer 360 use cases demand this full stack to generate executive briefings and account plans reliably
