Recommendation Engine for Broadcasters and Their Streaming Apps
A broadcaster has a schedule, an archive and an editorial voice to protect. TasteRay ranks the programmes you choose against what a viewer says, in your viewers' language.
TasteRay gives broadcasters a recommendation engine for their streaming apps. You send programmes that are playable today, including archive and catch-up, and one call ranks them against what a viewer says, with a written reason in the language you set. Editors stay in control of the list, and Enterprise adds custom AI, SLA and white-label.
Broadcasters serve a home audience in its own language, from a catalog shaped by rights windows and years of archive. Two questions come up first: can it speak to our viewers properly, and who decides what it may recommend? This page answers both, and shows a Polish response with our editing visible.
A Friday evening, in Polish
We wrote a ten-item list in Polish, mixing a news bulletin, a morning show, a children's cartoon, a documentary, a period adaptation, comedies and an educational series. The profile was a public television viewer who likes Polish productions and good comedy, and options.language was pl_PL. Both comedies led. The news bulletin and the morning show did not come back.
The reasons below say, in Polish, that the first title is a warm family comedy that asks no effort, that the second is a light romantic comedy of the right length, and that the third is a familiar classic. The sentences are the model's, unedited.
{
"vertical": "tv",
"context": {
"profile": "Widz publicznej telewizji, piątkowy wieczór, lubi polskie produkcje i dobrą komedię.",
"preferences": [
"polskie",
"lekkie",
"dla całej rodziny"
]
},
"items": [
{"id": "p7", "name": "Ranczo", "description": "Serial komediowy o życiu małej wsi.", "metadata": {"type": "series"}},
{"id": "p3", "name": "Zakochani po uszy", "description": "Polska komedia romantyczna.", "metadata": {"type": "film", "runtime_minutes": 104}},
// ...and 8 more: a news bulletin, a nature documentary, a children's cartoon and others
],
"options": {"count": 3, "language": "pl_PL"}
}meta.processing_time_ms 2,304. The model's sentences are verbatim.Ranczo
95 out of 100, taste_match_score
To jest dokładnie ten „bezpieczny port”, którego szukasz na piątkowy wieczór – polska komedia, która nie wymaga od Ciebie wysiłku, a która łączy całą rodzinę przy jednym stole dzięki wspólnym żartom.
- Twoja preferencja „dla całej rodziny” jest tu kluczowa. Humor jest uniwersalny, ciepły i wolny od toksycznych nawiązań – każdy w domu znajdzie w tym coś dla siebie.
Zakochani po uszy
92 out of 100, taste_match_score
Gdy chcesz komedii, która ma w sobie ten jeden, dodatkowy łyk ciepła, „Zakochani po uszy” są idealnym wyborem. To polska produkcja, która trzyma poziom humoru, nie obciążając Cię lękiem przed kiepskim kinem.
- Jest „lekkie” w sensie emocjonalnym. To film, który zostawia Cię z uśmiechem na twarzy, a nie z ciężarem w głowie. To definicja dobrze spędzonego wieczoru.
Czterej pancerni i pies
88 out of 100, taste_match_score
Szukasz czegoś, co Cię zabawi, ale jednocześnie da Ci poczucie, że oglądasz coś wartościowego i „prawdziwego”. Ten klasyk to bezpieczna przystań polskiej telewizji, która nigdy nie zawiedzie, niezależnie od tego, na jakiej jesteś fali humoru.
Fields shown: item.name, metadata.taste_match_score, explanation.why_match, explanation.key_factors.
What we left out of that response
Showing only the good sentences would mislead you, so here is what we removed. The key factors on the third result, a series about a wartime tank crew, treated it partly as a comedy or as light for everyone, which we would not stand behind. Several sentences on the first two results used odd figures of speech that a Polish reader would stop at. One contained a grammatical error, and another a misspelled verb.
For a broadcaster that is the main finding of this capture. The ranking was sensible, and the writing needs a native editor's review. Read a sample in every language you plan to ship, decide whether reasons appear on screen from day one, and consider why_match alone as a shorter, safer first release. Enterprise training on your content is scoped with you and is the place to raise voice and terminology.
What changes for a broadcaster
- Editors hold the list
- Your filter decides what may be recommended at all. Keep news, current affairs or anything sensitive out of the candidates, and the API only returns items from the list you send.
- Rights windows
- Catch-up and licensed programmes come and go by territory and date. Build candidates from your rights system at refresh time, and let cached rows expire before the window does.
- The archive
- An old programme with one line of synopsis gives the model little to reason from, unless the title is famous. Write a sentence or two per title before you test, since
descriptionis most of what it has. - A viewer's own language
options.languagesets the language of generated strings, so one integration can serve a regional audience. Our Polish request was also written in Polish, so test requests in the form you would really send.- Sign-in and profiles
- Many broadcaster apps are free to watch with light sign-in. A sentence or a theme button is enough for a ranked answer, so a viewer with no account still gets one. Keep kids profiles out of the integration unless your contract says otherwise.
For a broadcaster of your size
TasteRay Enterprise provides custom AI trained on your content, white-label delivery, an SLA, dedicated support, analytics and custom limits, and it can start with a pilot on one surface. No training data is needed to start. A DPA is available on request, and the security page sets out the data flow and the role of AI model providers.
You would deal with people who have years of experience running product, technology and growth at scale. More on the team page.
- Mike Wolniak, CEO
- Built and sold the VML hub in CEEMEA.
- Mike Jaskolski, CTO
- Led product and technology for Nokaut and Morizon, both of which went public. Builds the TasteRay engine, trained on more than 500,000 long-form reviews.
- Marius Labudda, Growth
- Drove user growth at Sportano and G2A.com.
- Kordian Klecha, Product
- Has built products and services used by millions.
When this is the wrong tool
This page is about what viewers see. Archive search for your own newsroom and production staff is a different product. If you want a platform across metadata, ad audiences and discovery, compare ThinkAnalytics. If you want your video analysed for mood and scene data, compare Vionlabs.
Programme names in the example are illustrative. We wrote the list ourselves, and TasteRay is not affiliated with any broadcaster named or implied by it.
Sources and further reading
Frequently Asked Questions
Can TasteRay write its recommendations in Polish or other languages?
Yes, options.language sets the language of generated text, and our Polish capture of 5 November 2026 came back in Polish. Quality needs checking per language: the ranking was sensible, but some sentences were clumsy and we left them out. Have a native editor read a sample before anything reaches viewers.
How does a broadcaster keep control of what gets recommended?
You choose the candidates. Filter out anything that should never be suggested, such as news or sensitive programming, then send the rest, and the API only returns items from the list you send. Editors can also write the themes themselves and read the reasons for each theme before a cached row goes live.
Does TasteRay understand old or little-known archive programmes?
Only as well as you describe them, unless the title is widely known. TasteRay does not watch your video. It works from each name and description you send, so a thin archive entry produces a thin reason. Write a sentence or two per programme, and let your editors judge the output.