1. Context & Environment (The Occasion Chart)
The Occasion Chart answers the fundamental question: "When will this bottle truly shine?" Our AI scans thousands of reviews for situational metadata, identifying recurring themes to match the spirit to the right moment.
Temporal Matching
Aligns spirits with time-based preferences—distinguishing between a "Sunny Afternoon" aperitif and a "Late Night" fireplace dram.
Social Calibration
Identifies the social "energy" of a bottle, ranging from "Quiet Contemplation" to high-energy "Celebrations."
2. Recommendation Intelligence
We map every spirit in a 300-dimensional vector space: the average word vector of all its review texts. The Cosine Similarity between those vectors gives three distinct discovery paths to grow your palate:
Closest Matches (The "Safe" Path)
High-overlap profiles with minimal vector distance. Ideal for finding "more of what you love" across different brands or regions.
Familiar Variations (The "Explorer" Path)
Foundationally similar but with a twist. These bottles share your core preferences while introducing new sensory complexities.
Expand Horizon (The "Challenge" Path)
Strategically chosen Polar Opposites: the least similar spirits, for calculated exploration without the risk of random guessing. Reviews of spirits share much of their language, so even opposites stay fairly close in this space; a match percentage would look high and mislead, so none is shown here.
3. Strategic Sensory Impact
Our engine enables a vertical palate trajectory, moving beyond simple "If you liked X, try Y" logic:
Occasion Pairing
Eliminate the "Right Bottle, Wrong Moment" syndrome. Data-driven insights ensure your choice matches the social context.
Palate Trajectory
Stop horizontal buying. Move through the world of spirits by following a sensory trajectory mapped by millions of data points.
4. Under the Hood
The models and software behind this page, and when they run.
-
fast-langdetect Google Cloud Translation
Language detection & translation: A review in another language is translated to English first, because the analysis models read English. The original text stays on the site.
When: Every new review, written here or collected from other sites.
-
spaCy
en_core_web_lgCharacter & occasions: Word lists read mouthfeel and structure (such as smooth or harsh) and the occasions people mention from the text. Community members set these themselves in the review form.
When: Every review collected from another site.
-
spaCy
en_core_web_lgscikit-learnSimilar bottles: Averages the word vectors of all reviews of a drink into one fingerprint and compares the fingerprints of all drinks (cosine similarity). No generative AI is involved.
When: When an admin refreshes the recommendations.
Deep Dives & Technical Documentation
NVIDIA: Recommendation Systems
How content-based filtering and deep learning predict consumer preferences at scale.
The Morning Paper: Word Vectors
Mapping the relationship between complex human concepts like "smoky" and "cozy."
Jay Alammar: Illustrated Word2Vec
A visual guide to understanding how embeddings create a searchable sensory space.