tastr

// your movie profile, read from your data.

tastr turns a watch-history export (Trakt, Letterboxd, or Bingers) into a full taste profile: who directs the films you love, which genres and decades you actually watch, how far you got with each series you started, how you compare to the crowd, a Spotify-Wrapped-style yearly recap, and a suggestions engine that recommends what to watch next — and what to watch to fill the gaps in your taste.

Everything runs local-first: your import never leaves the browser. It's parsed client-side, cached in localStorage/IndexedDB, and the only server-side call the app makes on your behalf is proxying TMDB requests (so your TMDB API key isn't exposed to the browser).

Features

Import

Drop the .zip export from one of three supported services and tastr detects the format automatically:

Source	What's read
Letterboxd	diary.csv / watched.csv (+ ratings.csv if present) — films only
Trakt	the JSON export (movie and episode history, show and episode ratings)
Bingers	watches.csv (+ ratings.csv if present), movies and episodes

Rewatches, ratings, and watch dates are normalized into a single internal format regardless of source. Episodes keep their season and episode number, so the series section can tell ten episodes apart from one episode watched ten times. Everything is stored in the browser — nothing is uploaded to a server.

Dashboard

• Top 10 directors and actors, ranked by a custom affinity score that combines how much you've watched someone with how far above your personal average you rate them, with Bayesian shrinkage so two five-star films can't outrank a director you've followed for a decade.
• Charts by decade and by genre, with a "you vs. the crowd" mode comparing your ratings against TMDB's public average.

Series

Television counted the way television works — in episodes, not titles. A show is one row but sixty viewings, so nothing here reuses the film maths:

• How far you got with everything you started: episodes seen against episodes aired, and a status that follows from it — finished, in progress, dropped (left unfinished for over a year) or sampled (three episodes or fewer).
• Time actually spent, episodes × runtime, rewatches included.
• A devotion score — log₂(1 + episodes) × quality × completion — where quality is the same Bayesian-shrunk distance from your own average used for directors, except a rating on the series counts as three episode ratings, because it judges the whole thing rather than one instalment.
• Genres ranked by episodes watched, not by number of shows: two long-runners outweigh six miniseries, which a count of titles gets backwards.
• Top creators and actors, weighted by the episodes you watched them in — a recurring lead and a guest star are not the same credit, a distinction that billing order makes for films and episode count makes here.

Anything that needs TMDB (episode totals, runtimes, genres, cast) degrades to a visible "unanalyzed" instead of guessing.

Wrapped

An annual recap, picked per watch-year: eight cards (how you rate old vs. new movies, how deep you dig into genres/directors, how you compare to the crowd, your rhythm across the year, how much you rewatch...) each resolving to a trait, plus an overall verdict and a shareable image export.

Suggestions

Two lists built from your taste profile:

• For you — unseen films that match what you watch and rate well, sourced from TMDB recommendations on your favorite films, filmographies of directors/actors you love, and the best of your top genres. Ranked by taste × crowd consensus, with the crowd signal weighted by how much your ratings actually agree with the public.
• Blind spots — genres, decades, and director filmographies you've barely touched, each paired with the most canonical film to close the gap.

Choose all time (recent viewings weighted roughly 2× a viewing from three years back) or recent tastes (last 3 years only) as the basis for the profile.

Getting started

1. Install dependencies

pnpm install

2. Configure TMDB

Most features (credits, genres, ratings comparison, suggestions) need a TMDB API key. Create .env.local:

# either a v3 API key (32 hex chars) or a v4 Read Access Token — both work
TMDB_API_KEY=your_key_here

Without it, import and basic browsing still work; anything needing movie metadata shows a "TMDB not configured" notice instead of failing.

3. Run the dev server

pnpm dev

Open http://localhost:3000. You'll be redirected to your browser's preferred locale (en or it).

4. Try it

Go to Upload, drop a .zip export from Letterboxd, Trakt, or Bingers, then check out Dashboard, Series, Wrapped, and Suggestions. (A Letterboxd export has no episodes in it, so the Series section will tell you it found none.)

Scripts

pnpm dev      # start the dev server
pnpm build    # production build
pnpm start    # run the production build
pnpm lint     # eslint

Tech stack

• Next.js 16 (App Router, React 19)
• TypeScript, strict mode
• Tailwind CSS 4
• IndexedDB for the TMDB credits/suggestions cache (films and series in separate stores — TMDB numbers the two independently), localStorage for the active import
• No backend, no database — the only server code is the /api/tmdb/* proxy route that attaches your API key to outbound TMDB requests

여기까지가 README 앞부분입니다. 나머지는 원문에서 읽을 수 있습니다. README 전문 보기 ↗