How Lumora decides what to suggest next
Three taps under a title, a watchlist, and how far you got: the signals the recommender reads, and why every profile gets a different list.
Lumora Editorial
· 3 min read · Inside Lumora
Every recommendation engine makes a claim about you. This one makes a small one: it reads what you do here, and nothing else. No import of your viewing history from another service, no profile of you assembled somewhere else and applied to this catalogue.
The taps are the clearest signal
Open any title and scroll to the overview. Under the facts there is a row of three controls: Like, Not for me, Not interested.
They are not three shades of the same answer:
- Like tells the engine this belongs in the pool your shelves are drawn from.
- Not for me records that this particular title missed — without condemning the genre it came from. One bad film is not a bad decade of them.
- Not interested keeps the title itself out of your suggestions from then on.
That distinction is the whole reason there are three buttons instead of a star rating. A rating tells you a number. These tell the engine why, and the why is what survives into the next recommendation.
What else it reads
Taps are the honest signal, but they are also the rarest — most sessions end without anyone pressing anything. So the engine also reads behaviour you produce just by using the site:
- What you opened and how far you got. A title abandoned at four minutes matters differently from one finished at eleven at night.
- Your watchlist. Adding something is intent that outlives the session.
- Genre overlap between what you have engaged with and what is in the catalogue.
What it does not read: anything from outside Lumora, anyone else's activity, and any profile other than the one you are using.
Why every profile gets its own list
Profiles on Lumora are not a UI filter. Each one has its own history, its own watchlist and its own feedback, stored separately and read separately, so two people sharing an account end up with genuinely different catalogues.
A kids profile is stricter than that. Titles are filtered in the pool the engine draws from and again at the moment a row is assembled — because a filter applied once is a filter that a second code path can forget.
What it deliberately will not do
A recommendation is a guess you get to correct. The moment it becomes a box you cannot climb out of, it has stopped being useful.
So the engine has limits worth stating plainly:
- Cold start is honest. With no signals, it shows what everyone sees — trending, new arrivals, genre rows — rather than inventing a personality for you.
- It reconsiders. Scores are held briefly and rebuilt, so an evening of one genre does not define the next month.
- Popular is not the same as good for you. Shared rows fill the page, but once your own signals exist they are ordered by what fits, not by what everyone else watched.
In this story
Seeing it work
The fastest way to watch the mechanism move: open the catalogue, find something you actually liked, and press Like. Go back to Discover. The rows at the top are yours now, and each one is labelled with the reason it is there — a title you enjoyed, a genre you keep returning to.
If the list is wrong, the row of three buttons is the correction. That is the entire interface: no settings page, no onboarding questionnaire, no interest checkboxes. Just the answer, given one film at a time.
- Recommendations
- Profiles





