TasteCheckout

Inspiration

Gift recommendations usually flatten people into demographics and bestseller lists. TasteCheckout starts from a better question: what cultural constellation makes a gift feel recognizably theirs? Qloo makes those cross-domain connections computable.

What it does

A user provides 2–6 cultural interests, relationship, occasion, budget, exclusions, and optional context. TasteCheckout uses Qloo to resolve those signals into cross-domain affinities, builds an explainable gift-search brief, discovers candidates, ranks them against budget and exclusions, and shows why each item fits. A human must select and explicitly approve an item before any checkout step.

How we built it

The Node.js agent uses the official Qloo Hackathon Kit through bounded qloo exec recommend workflows with explainability enabled. Qloo provenance is retained through ranking and displayed in the agent trace. Channel3 is an optional discovery adapter; a deterministic catalog keeps the project reproducible. PayPal Sandbox is isolated behind a separate server-side approval boundary. The browser never receives API keys.

The repository includes contract tests, integration tests, four deterministic evaluation scenarios, security documentation, and a Node 22.19 container. Demo mode is clearly labelled and never pretends that a live Qloo request occurred.

Challenges we ran into

The hardest part was preserving useful Qloo evidence across multiple agent steps instead of reducing the result to opaque product scores. We also designed the live harness as a timeout- and output-bounded subprocess, kept all credentials server-side, and prevented catalog discovery from silently becoming checkout authority.

Accomplishments that we're proud of

• Qloo is functionally central rather than decorative.
• Recommendations stay within budget and retain an explainable cultural affinity.
• Human approval and payer approval are separate, enforced checkpoints.
• 10 automated tests and all 4 evaluation scenarios pass.
• The project is open source under MIT with reproducible setup.

What we learned

Taste is more useful as a network of affinities than as a list of keywords. Agent traces also become much more trustworthy when every recommendation can point back to the cultural signal that caused it to rank.

What's next for TasteCheckout

Activate the event key when Qloo delivers it, run and document the live smoke test, add redacted Qloo evidence, and expand the catalog adapter while keeping Qloo central and checkout explicitly human-controlled.

Built With

• qloo