Why Qloo
What only Qloo's taste graph gives Vasal, which of its capabilities the code uses, and what breaks without it.
A podcast host does not need a list of famous people. They need people whose fans already listen to shows like theirs. That answer lives in audience data, not in a model's memory, and Qloo is the source Vasal uses for it.
What only Qloo gives this product
Qloo's taste graph connects entities of different kinds through the audiences they share: podcasts, people, films, TV shows, books and more. Vasal uses four things that no podcast directory and no language model holds.
- Cross-domain taste: given a podcast, Qloo can say which people its listeners love. The link runs from a show to people, so a guest can come from music, film, sport or anywhere else the audience's taste reaches. A directory of podcasts cannot answer that.
- How strong the match is: Qloo returns an affinity score for a person given a show. Vasal asks for all of a board's guests in one call, so their scores sit on one scale and rank against each other.
- What the guest brings: Qloo reports who a person's fans are, by age band and gender. Vasal compares that with the show's own audience to find a group the guest's fans lean toward more than the show's listeners do.
- What the two audiences share: Qloo compares two audiences and returns the tags they have in common. Vasal turns the best few into the "shared tastes" on a board, and into material for a brief.
Qloo also supplies the checks that keep a board realistic. A person with a death date is removed, and a person whose popularity is at the very top of Qloo's scale is labelled a household name.
The capabilities the code uses
| What Vasal asks | Qloo capability | Where it is used |
|---|---|---|
| Which Qloo podcast is this show | Entity search, podcasts | Finding the show |
| Which people this show's listeners love | Insights, people, with the show as the signal | Research |
| Who a person's fans are | Insights, demographics | Research and the "leans toward" group |
| How well a person fits a show | Insights, people, with the show as the signal and the person as the result filter | The fit on every board |
| What two audiences share | Analysis, compare | Shared tastes |
| Which films, shows and books are coming | Insights, films, TV shows and books, with a release year filter | Research |
| Who a name is | Entity search, people, and entity lookup, through Qloo's own harness | Research and Without Qloo |
| Whether interest in a person is rising | Trending, people | Switched off by default |
Every Qloo call shows the request for each one, with every parameter name.
What breaks without it
Take Qloo out and the same model still suggests guests, but from memory alone. It tends to name the people it has read about most, which are the well-known ones. A model has no listener data, so it cannot tell a guest whose fans match your audience from a guest who is only famous. It also cannot check a name against the show's whole episode history, so it can suggest someone your show had on years ago.
Vasal does not leave that as a claim. Every board has a Without Qloo tab. It asks the same model for guests with no Qloo data, then lets Qloo name and score each guess and checks each one against the show's episode history. The home page totals how many of those guesses had already been on the show or hosted it.
The backtest tests the other side, which is whether Qloo's fit score means anything for audience size. Across 755 episodes of 8 large podcasts, guests with a higher fit drew more views, after allowing for how long each episode had been online and for the guest's general fame. The link is small and positive.
What the code does with it
Qloo supplies the data and the code does the arithmetic. The AI model chooses which questions to ask and writes the prose, and it never writes a number. That split is why Vasal can be checked: every figure on a board traces to a Qloo call, and How it works shows where each one is computed.