Willie — Wine Recommendation Engine

Challenge

Generic "you might also like" recommendations miss the mark for wine buyers who care about origin, vintage, and the story behind a bottle. The retailer needed personalisation that explains itself — recommendations the customer can understand and trust without needing a sommelier.

What we built

We built a recommendation pipeline on BigQuery and dbt. The wine catalogue and order history live in BigQuery; dbt transforms them into the feature tables that the recommendation engine reads. The engine surfaces personalised wines with readable explanations: geographic match (same region as a wine they bought), vintage year overlap (birth year, anniversary), or name overlap (playful thematic match). Each recommendation comes with a one-sentence reason. A Flask backend serves recommendations via a small API that embeds into the retailer's site.

Result

Explainable recommendations that go beyond "the algorithm said so." Customers can read the reason for each recommendation out loud — geography, a personal year, a name connection. The BigQuery and dbt backbone is in production; the front-end embed is the active build target.

Tech used

• BigQuery
• dbt (data build tool)
• Python / Flask API
• Geographic + vintage + name similarity matching
• Explanation generation per recommendation