Sightlines · about
Sightlines is an atlas of about 3,400 films. We read every film for what it is about and how it feels, trace what it descends from and what it shaped, and file it in the viewing threads it belongs to. Then we use all of that to help you find films: by mood, by a film you loved, or by bringing your own list. It is a personal project, not a database dump. Every film was chosen, written up and wired into the web by hand.
Sightlines is a curated collection, not everything ever made. A film gets in for one of five reasons:
Recent releases join as they land. If a film you look for isn't here yet, the search says so, and every title people ask for goes on the list to add.
There are five pages, and they all draw on the same readings. Browse is the whole collection on one grid. Sort it by tone, narrow it by decade, genre, mood or place, or group it by viewing thread. Ask takes a sentence: what you're in the mood for, a film you loved, or a question like “What films led up to Sinners?” It answers with a grid of films and says why for each one. Your Films reads a list you bring, from Letterboxd, a screenshot, a CSV or typed titles. It groups your films into viewing threads, reads your taste, finds their kindred films and lines of sight, and writes a course. Essays follow one idea through a century of film. Courses are guided, spoiler-free runs through a set of films.
Every film has its own page: what it's about and how it feels, its dossier, its lineage with how the influence works, its kindred and sideways films, the theory and reception behind it, and the essays, courses and viewing threads that touch it.
New here? How it works → walks through every page.
At the heart of the system is a dossier written for every film against a fixed template: industry and production, technology, technique, narrative, genre, authorship, movement, era, themes and reception. Because each dossier covers every dimension, we can re-extract the facets as the vocabulary grows without rewriting a card.
The facets you browse by come from two layers. A rule layer resolves curated facets from signals already attached to each film: keywords and genres, the director, the original language, and craft markers from the dossiers. On top of that, language-model passes read each film's plot and dossier to infer what keywords miss, that a film is about justice, or addiction, or the atomic age, and to fold thousands of fragmentary keywords into a few dozen coherent dials. The lineage graph is built the same way, from specific craft debts between films, not vague resemblance.
Under the hood it's plain Python and static HTML and JavaScript. The live parts run on one small server: the search behind Ask, the reading of your films, the course writer. Everything else is built ahead of time. Film metadata and posters come from TMDB.
Facets are the dials you browse by: mood, subject and theme, setting, character, craft, movement. Two things are worth knowing.
First, a facet folds related keywords into one idea. Rather than expose the thousands of raw keywords on each film, Sightlines gathers the tightly related ones into a single dial, like “Drugs & addiction,” “Nuclear & the atomic age” or “Surveillance & social control.” One dial absorbs its synonyms, so you browse by the idea rather than by fifty overlapping tags.
Second, facets are deliberately loose. They're inferred, by rules and by language-model passes reading each film, not stamped one film at a time. A film is rarely about one thing, so a facet marks a film that substantially belongs, and reasonable people will disagree at the edges. That looseness is the point. It's what lets a search for “justice” or “memory” or “the atomic age” surface films you'd never have filed there yourself. Treat facets as doors into the collection, not a catalog.
Beyond its facets, every film carries a reading through the film philosophy of Gilles Deleuze, the French thinker who built a whole system in his two Cinema books for describing what film does with movement and time. It rests on two big ideas. The movement-image is classical cinema, organized around action: perception, then feeling, then deed. The time-image is modern cinema, from Neorealism on, where time itself surfaces and characters watch rather than act. Beneath those sit dozens of finer concepts: the affection-image and the close-up, the crystal-image, the seer, the powers of the false.
On Sightlines those concepts become a second set of facets. Each is a door to the films that embody it, every concept has its own page, and all are defined in plain language in the glossary. You don't need to know Deleuze to enjoy them. They're written to make you notice how a film thinks, not to impress. Each reading works as a method of seeing rather than a theory attached afterward: start here, look at this, notice where the film breaks a pairing, then go looking for the same relationship elsewhere in it. Curated runs of these films are gathered as guided, spoiler-free courses.
Sightlines was designed, written and built by Adrian Chan, a service and AI-interaction designer, at gravity7.com, where he also curates Inquiring Lines, a library of AI and LLM research. The semantic-search and recommendation methods documented there are the ones Sightlines applies to film. It's an ongoing experiment in tracing how influence travels through film over time. If it sends you toward a film you'd never have found, it's doing its job.