Introduction
What TasteCase does, who it serves and what it doesn’t claim.
TasteCase helps teams test an AI assistant that recommends restaurants. It finds cases worth checking, lets you decide what a good answer should do, and saves those checks for the next time you change your assistant.
Connect your assistant → find a case → review it → save a test → compare versions.
What TasteCase does
Imagine two fictional diners visiting the same city. They have the same practical needs, but different interests: one listens to Radiohead, the other to Taylor Swift. Should your assistant give them the same restaurant suggestions? Maybe. TasteCase helps you inspect that question rather than guessing the answer.
You supply a list of restaurants and the dinner request. TasteCase asks your assistant about both diners and saves the answers beside Qloo’s rankings. You review what happened and decide whether to keep the case as a test.
Later, compare your current assistant with a changed version using those same saved tests. The two versions can use the same model with different prompts. You do not need two different models.
Who it’s for
Developers and testers building AI travel or dining concierges. TasteCase tests their assistant; it is not a restaurant-booking service for travellers.
Where Qloo helps
Qloo connects public cultural interests, such as music and films, with places. It helps TasteCase find restaurant pairs worth investigating. It does not decide what a particular person should like.
Start with Inspect a case. No account or model key is needed to read the saved example.
What it can’t tell you
- A different ranking is not proof that your assistant is wrong.
- If Qloo has no evidence for a restaurant, that does not mean the diner dislikes it.
- An incomplete activity record cannot prove your assistant skipped a tool.
- We have not yet measured whether this finds more useful cases than ordinary test generation.