How Envie
works.
A taste profile is a mathematical representation of your visual preferences, built automatically from the items you like and dislike.
Unlike a wishlist or a saved collection, a taste profile captures the underlying pattern — not just what you liked, but why those items cluster together aesthetically.
What a taste profile is — and why it beats browsing
A taste profile captures not just the individual items you like, but the underlying aesthetic patterns that connect them: a recurring preference for clean silhouettes, warm metals, or structured shapes. Traditional browsing forces you to consciously identify and search for what you want, which only works once you already know your taste. A taste profile discovers it for you by learning from patterns in what you are drawn to, including things you might not yet be able to articulate.
How swiping builds a preference signal
Each swipe sends a directional signal: a like moves your profile toward that item's aesthetic cluster; a nope pushes it away. Recent swipes carry more weight than older ones — the system uses exponential decay so your current preferences count more than what you liked six months ago. The profile stabilises when your swipes become consistent enough that the system can reliably predict what you will like next. That is when calibration completes and your personalised feed begins.
How the embedding model understands visual style
Envie maps every item to a point in a 768-dimensional space of visual style, using a model trained on fashion imagery. Items that share aesthetic qualities — similar colour palette, material, silhouette, or weight — cluster together in this space, regardless of brand or category label. A sculptural gold ring and a minimal gold band land close together; a chunky silver ring lands further away. When you like something, your profile moves toward that cluster. When you nope, it moves away.
How recommendations get more accurate over time
The recommendation pipeline retrieves candidates by searching the catalogue for items close to your current taste profile, then scores each one against your full swipe history — measuring similarity to liked items and dissimilarity to noped items simultaneously. A sampling layer adds calibrated exploration, giving items your profile is uncertain about a fair chance to appear. As your swipe history grows, both retrieval and scoring become more precise: the system has more signal to work with, and your profile vector stabilises closer to your true aesthetic.
How Envie differs from Pinterest and generic shopping
Pinterest relies on conscious curation: you save things you already know you like, and the algorithm amplifies that existing signal. Envie learns from passive signals — the cumulative pattern of swipes, not deliberate selections — and incorporates negative feedback that most platforms discard. It does not surface items because they are popular or trending. It does not follow social graphs. Every recommendation comes solely from your individual swipe history, which means two users in the same city with similar demographics can have entirely different feeds.
Common questions
How does taste profiling work?
Every item on Envie has a 768-dimensional embedding — a mathematical fingerprint of its visual and semantic style, generated by a model trained specifically on fashion imagery. When you swipe, Envie plots your likes and nopes in that embedding space and builds a personal taste vector: a point in style space that represents your aesthetic. This vector is updated with every swipe, with exponential recency weighting so your current taste counts more than what you liked months ago. The result is a living model of your preferences, not a static snapshot.
How does Envie learn my style?
Envie begins with a calibration phase. You swipe through a curated set of seed items chosen to maximally explore the range of styles in a category. The system tracks not just the direction of your swipes but their consistency: once your last ten swipes align with the profile built from earlier swipes at 80% accuracy, calibration is complete and your personalised feed begins. From that point, every swipe adjusts your taste vector — likes attract, nopes repel — and the feed reshapes accordingly. The profile never freezes; it continues updating as your taste evolves.
What makes Envie different from Pinterest?
Pinterest is a manual curation tool: you consciously save things you want to remember, and the feed surfaces more of what you have already saved. Envie is a discovery engine that surfaces things you did not know you wanted by learning the patterns in your swipes, including negative signals that most platforms ignore entirely. Envie does not track popularity, trends, or social signals. It uses a Determinantal Point Process to balance exploitation — showing you more of what you love — with exploration, deliberately introducing items just outside your comfort zone to expand your taste rather than lock you into it.
How does the recommendation algorithm work?
The feed pipeline runs in three stages. First, an approximate nearest-neighbour search retrieves around 500 candidates from the item catalogue using multiple probe vectors derived from your liked items, ensuring diverse coverage across your taste space. Second, each candidate is scored: a weighted attraction score (similarity to liked items, with recency weighting) minus a repulsion score (similarity to noped items). Third, Thompson sampling adds calibrated exploration — items your profile is uncertain about receive a probability boost to appear. The final batch of 20 cards is assembled using diversity selection, so you never see 20 near-identical items in a row.
How do I discover my personal style?
Personal style is not chosen consciously — it is revealed through repeated visual choices. Most people struggle to define their taste because they have never had a systematic way to collect and analyse those choices. Envie accelerates the process by giving you a fast, low-friction signal loop: swipe on items, observe what you are consistently drawn to, and let the pattern surface. After calibration, your taste profile is a concrete model of your aesthetic that you can use as a reference — not just inside Envie, but when shopping anywhere. The glossary at envie-app.com/glossary also covers the terminology used to describe different style aesthetics.
Ready to train your taste?
Request early access