Most apps that promise "recommendations" really surface what's popular or what people-like-you watched. The best app for recommendations from your taste is one that models your actual watch history — and that's what Intertitle does: it reads your Letterboxd history and recommends tonight's film from your actual taste, not the algorithm's guess.
TL;DR
Streaming services recommend what keeps their subscribers engaged. Lists titled "best movies of 2024" recommend what critics or crowds ranked highest. Neither approach looks at your specific history — the directors you revisit, the eras you gravitate toward, the films you rated five stars at 1 a.m. Intertitle does. It builds a taste profile from your ratings and watch history, then ranks unwatched films by how closely they match it. The recommendation isn't a crowd consensus; it's a fit score against your own taste.
Popularity vs personal taste
Every recommender system makes a foundational choice: optimize for the crowd, or optimize for the individual.
Streaming platforms lean on collaborative filtering — grouping users by viewing behavior and surfacing what similar groups watched next. That approach works well at scale and keeps aggregate engagement high, but it regresses toward the popular. If you've spent years developing a precise taste for slow-burn psychological dramas from the 1970s, the collaborative signal from millions of users who mostly watched action blockbusters will dilute that signal considerably. The result is a recommendation queue that looks plausible but rarely feels right.
Generic "top movies" lists — whether editorial or algorithmically ranked — have the same problem from a different angle. They optimize for consensus: critical acclaim, box-office reach, or viral social traction. A film near the top of every "must-watch" list may be genuinely great and still wrong for how you watch.
The distinction matters because taste is specific. You don't want a film that's broadly good; you want the film that fits you tonight. That requires a profile built from your own data, not an approximation of similar users or an aggregate of critical opinion.
How the three approaches compare
| Streaming built-in recs | Generic "top movies" lists | Intertitle | |
|---|---|---|---|
| Basis for recommendations | Popularity and collaborative filtering (what similar users watched) | Critical consensus, trending, or crowd rankings | Your personal taste profile built from your own ratings |
| Uses your history | Partially — viewing activity within that platform only | No | Yes — full watch history and ratings |
| Result shape | An endless, popularity-ordered feed | A static consensus ranking | A short shortlist ranked from your taste, best match headlined |
| Free | Included with a paid subscription | Yes (usually) | Yes |
| Platforms | Platform-specific apps | Web, various | iOS, web |
How Intertitle builds your taste profile
Rather than grouping you with similar users, Intertitle encodes each film as a vector in a high-dimensional space where proximity represents aesthetic similarity — tone, pacing, decade, directorial style, narrative structure. Your ratings become weights: films you loved pull the model toward the parts of that space they occupy; films you disliked push it away. The result is a taste profile that reflects your actual aesthetic, not a bucket labeled "fans of genre X."
When you ask for a recommendation, Intertitle scores every unwatched film against your profile and returns a short, taste-ranked programme led by your best match. There's no quiz, no genre picker, no trending row. Just your own history, turned into tonight's programme.
The more you've rated — in Letterboxd or directly in Intertitle — the sharper the profile becomes. If you're starting from a Letterboxd export, you may already have hundreds of rated films, which gives Intertitle enough signal to return precise recommendations immediately.
For a deeper look at the mechanics, read how taste embeddings work.