Math you can read
Recommendations are a weighted dot product over your tag preferences. No neural net. The formula is right there in the docs.
v0.1 — invite-only alphaRecommendations
Recommendations
that show their work.
Likewyse is a TV recommender powered by a deterministic scoring engine. Swipe through shows, the math learns from every right-swipe. No black-box rankings. No LLMs deciding for you.
- Catalog
- 5,000
- shows seeded
- Threshold
- 20
- swipes to recs
- LLMs in path
- 0
- deterministic only
House of the Dragon
2022FantasyDrama
Succession
2018DramaSatire
Score 0.84
The Last of Us
2023DramaPost-apoc
How it works
Three steps. No magic.
Just a swipe loop and a transparent formula. You could pull out a calculator and check the recommendations yourself.
01
Swipe
Right for love, left for pass, up for "haven't seen it." A full-viewport deck of TV shows, decided one card at a time.
02
The math runs
Every right-swipe contributes to a transparent tag-weight vector. Actors, creators, themes, networks — each signal has its own pull.
03
Recommendations land
Top 5 shows by weighted dot product. Scored, ranked, deterministic — recomputed on every visit, never cached, never decayed.
Why Likewyse
A recommender with its
cards on the table.
No LLMs in the recommendation path
The thing deciding what to show you is a transparent algorithm, not a chatbot. Auditable, predictable, and offline-capable.
Tastes expand, not invalidate
A like from 2024 still counts in 2030. Weights never decay. Your taste history is honored, not silently aged out.
Integers everywhere
No "0.7321 affinity score" in the UI. Preferences are shown as whole numbers, top to bottom — math in the engine, plain numbers for you.
Build your taste profile.
Twenty swipes is all it takes. The engine takes it from there — and you can audit the math at any point.