PALATE

Inspiration

Generic assistants can plan an evening, but their recommendations often ignore taste. PALATE asks a concrete question: what should I do tonight? It uses cultural affinity across food, film, music, and place to connect an itinerary to the things a person already loves.

What it does

Give PALATE a city, a mood, and a few favorite things. Its agent resolves references, grounds the mood in tags, asks for candidates for dinner, film, music, and a nightcap, then critiques each shortlist. When a slot is weak, it changes the signals and queries again. The result includes an itinerary, confidence, taste connections, alternatives, and the request behind each choice. The interface exposes the full plan, query, critique, and revision trace.

How I built it

The Python standard-library server connects a small, explicit agent loop to a Qloo client. The live client is implemented for the hackathon API's /search, /v2/tags, and /v2/insights endpoints, with the key kept server-side. A separate deterministic client uses a bundled synthetic taste graph so development and tests can run without a key. The UI is vanilla HTML, CSS, and JavaScript. The repository is MIT licensed and has 62 passing unit tests, including Qloo request-contract tests using an injected fake transport.

Challenges and lessons

The hackathon API key is provisioned separately, so I kept simulated and live sources distinct. The UI labels simulated results and any fallback plainly; a synthetic recommendation is never presented as a live Qloo result. Building an inspectable critique-and-revise loop showed how candidate count, matching taste threads, and repeated picks can trigger a useful second query instead of a confident but weak first answer. The live Qloo path has not yet been verified end-to-end with a provisioned key; the current demo should be evaluated as a clearly labeled simulation until that check succeeds.

What's next

Run the live integration with the hackathon key, tune requests against real results, and surface Qloo explainability metadata directly on each recommendation card. Then expand the planner to group tastes and other kinds of outings.

Source and setup: https://github.com/bgrubbs1/palate-qloo-agent

Built With

• python
• qloo