Choosing a movie shouldn’t feel like homework. We wanted something fast, fun, and addictive. SwipeFlix was inspired by the idea that preferences can be learned extremely quickly if the system is reactive enough.
So we built a recommender that adapts every time you swipe.

SwipeFlix is an intelligent movie-recommendation app that works like Tinder for movies.
Users swipe like or dislike, and the system instantly updates its understanding of their taste.

It does this by:

• Tracking liked genres, actors, countries, and decades
• Computing a probability score using feature ratios (likes/dislikes)
• Picking new movies using a weighted scoring system
• Mixing 90% smart picks with 10% exploration, where exploration starts at 50% and decays by 5% per movie
• Generating insights like “You seem to like 90s crime movies starring Brad Pitt”
• Showing a dynamic likelihood bar for how much you’ll enjoy the next movie

Over time, the recommendations become sharper, more personalized, and surprisingly accurate.

The frontend was built using HTML, CSS, JavaScript, and TailwindCSS.

The backend was built using Flask, and the core recommendation engine is a Bayesian-style scoring model that transforms the like to dislike ratio of movie features into a probability estimate using a logistic function. This converts the user’s historical like/dislike behavior into a continuous probability from 0 to 1.

• Ensuring the probability system stayed stable when data was sparse
• Handling movies with features the user had never rated before
• Preventing the system from getting stuck in a genre “feedback loop”
• Designing an exploration system that’s both smart and unpredictable
• Managing the removal and cycling of movies from the recommendation pool
• Making the learning updates fast enough to feel instant during swipes

• We built a fully functioning adaptive recommender from scratch
• The probability model actually behaves realistically and self-corrects
• Insights like “You seem to like…” turned out surprisingly accurate
• The explore–exploit balance (50% → 10% decay) created natural diversity
• It genuinely feels fun to swipe through movies and watch it learn

• Logistic models are fantastic for real-time personalization
• Small weights (like 0.5 for country) dramatically affect recommendation stability
• Exploration is essential, or the system becomes too confident too early
• Users respond really well to explanations (“You seem to like…”)
• Building a good recommender = balancing stats, intuition, and UX

• Add a “why this movie” explanation for transparency
• Build persistent user profiles and long-term preference memory
• Introduce collaborative filtering to blend personal + global trends
• Add trailer previews, watchlists, and social features
• Release a mobile app version for smoother swiping