Hi Brahma,

Really impressive work! The problem framing alone sets this apart — stockouts vs. waste as two sides of the same inventory problem is exactly the kind of clarity that makes a solution easy to champion internally at a store level.

A few things I genuinely liked:

The decision to route natural language questions to pre-approved Databricks views rather than letting an LLM write raw SQL is a smart, production-minded guardrail. A lot of AI demos skip that entirely.

The Inventory Story page is an underrated feature. Translating Gold layer metrics into plain-language narrative for business users is harder to do well than it looks, and it makes the whole solution accessible beyond the data team.

Using the Gold layer to power four different surfaces (dashboard, product actions, story, and the AI assistant) shows real architectural thinking about reusability.

One thing I'd love to understand better — how does the question-to-view mapping work under the hood? Is it semantic similarity, keyword routing, or LLM function calling? That feels like the most interesting technical piece and the write-up leaves it a bit open.

Great submission overall. This is the kind of practical, well-scoped Data + AI application the community needs more of. Good luck!