Letterboxd is where your history lives — the diary, the ratings, the four-star films that stack up over years. What it doesn't do is turn that history into a recommendation engine. There's no personalized shortlist for tonight, no score that tells you how well an unseen film fits what you actually love.
Intertitle is that missing piece. It reads your Letterboxd data and ranks unwatched films by how closely they match your taste — a proper recommendation engine built on top of the history you've already built.
What the engine actually does
When you import your Letterboxd history, Intertitle identifies the films you've rated four stars or higher — your loved films. Each of those films sits somewhere in a high-dimensional embedding space that captures the kinds of cinematic qualities that made them feel right to you: the tone, the texture, the sensibility.
Every unseen film in the catalog has an embedding too. The recommender measures how close each candidate sits to your loved films in that space — a cosine distance computed against the nearest loved films in your profile. The closer the match, the higher the score. The score you see (0 to 1) is taste affinity: 1 minus that cosine distance, where a higher number means a closer fit. A 0 means the engine has nothing to work from yet.
The ranking is grounded in your history — not a random shuffle, not an editorial pick. The films you're shown are drawn from the ones the geometry says fit your taste best. For more on how the embedding space is constructed and what it captures, see how taste embeddings work.
What Letterboxd does (and what it doesn't)
Letterboxd is the best film diary on the market, and it's a social network built on that diary. You follow friends, read reviews, browse themed lists, and see what the community is watching. Discovery is film-by-film (similar titles on each film's page) or social (trending in your network) — useful, but not personalized to your specific taste profile.
There's no Letterboxd feature that reads your diary and generates tonight's shortlist ranked by match. That gap isn't a criticism — it's a different product. Letterboxd is where your history lives; Intertitle is where that history generates something actionable.
The two tools work together: keep logging in Letterboxd, connect your handle in Intertitle, and the sync keeps your taste profile current as you add new films.
Why popularity isn't the right signal
Most streaming recommendation engines are built around what's popular — what a large group of similar users watched, what's trending this week, what the platform wants promoted. Popularity is a reasonable heuristic when you know nothing about a viewer. It becomes a bad one the moment you have actual taste data.
Intertitle's matching signal is your loved films, not aggregate viewing behavior. A film from 1974 that sits close to your taste profile ranks ahead of a widely-watched release from last month if the geometry says so. The engine doesn't treat popularity as a tie-breaker; the score is the taste distance, nothing else. If your taste skews toward slow-burn arthouse or 1970s genre cinema, a taste-distance engine finds those films — a popularity engine doesn't.
For a deeper look at how this differs from collaborative filtering, see how taste embeddings work.
The cold-start limit
The engine requires taste data to work. Specifically, it needs films you've rated four stars or higher — those loved films anchor the taste vector it matches against. If you've rated very few films, or none of them reach that threshold, the profile is underspecified and early picks will be broad rather than precise.
This is an honest constraint, not a limitation to design around with a genre quiz or an onboarding survey. The right answer is to rate films — in Intertitle directly, or by logging in Letterboxd and letting the sync pull them in. Importing from Letterboxd is the fastest way to warm the profile if you have an existing diary.
Every film you add sharpens the next programme. The match score on each recommendation reflects exactly how much confidence the profile currently has — 0 means cold start; anything above that means the embedding geometry found something real.
Reading your match score
The number next to each recommendation is a taste affinity score from 0 to 1. It's computed the same way for every surface in Intertitle — Tonight, your watchlist, and the browse library all use the same signal, so a 0.87 in the browse view means the same thing as a 0.87 on tonight's bill. The match score explained walks through what the number means in practice and how to use it.
You can also see how your taste is distributed across genres, directors, and eras in the Taste Map — it's a useful way to understand what the recommender is working from.
Getting started
If you have a Letterboxd account, the fastest path is a data export: Settings → Import & Export → Export Your Data. Intertitle reads the ZIP and builds your profile on import; connect your handle once and subsequent syncs are automatic. A step-by-step walkthrough is at getting recommendations from your Letterboxd.
If you don't have Letterboxd history, rate films directly in Intertitle — the engine works the same either way. The more loved films in your profile, the sharper the recommendations get.