Pulse Taste Scout

• 

  Pulse Taste Scout — Qloo-powered cultural intelligence for explainable cross-domain recommendations.

Inspiration

AI agents can reason, plan, search, and execute — but when they need to understand taste, they often fall back to generic popularity, proximity, or an LLM's best guess.

Pulse Taste Scout was built around a simple question:

What changes when an agent can reason with a real cultural affinity graph instead of guessing what someone might like?

What it does

Pulse Taste Scout turns a small set of cultural preferences into an explainable, cross-domain recommendation.

A user provides a few taste seeds, a city or context, and a real decision they are trying to make.

The agent then:

• resolves the user's taste seeds into real Qloo entities
• queries Qloo for cross-domain cultural affinities
• builds a candidate set from Qloo results
• evaluates whether the evidence is strong enough
• performs one bounded retry when evidence is sparse
• ranks the strongest candidates
• explains why each recommendation fits
• shows confidence and limitations

The result is not just a list of places. It is an evidence-bound recommendation showing how the user's cultural signals influenced the decision.

How we built it

Qloo is the core intelligence layer.

The application uses Qloo to resolve free-form taste seeds into real entities and to retrieve cross-domain affinity signals through the Insights API.

The agent workflow has six stages:

1. Intake
2. Entity resolution
3. Taste graph query
4. Evidence critic
5. Ranking
6. Explanation

The app also includes a visible counterfactual:

Qloo enabled vs. Generic baseline

The generic baseline can only suggest broad categories. The Qloo-powered path returns named entities ranked from the user's resolved cultural signals.

Without Qloo, the product becomes a generic recommendation interface. With Qloo, it becomes a culturally grounded decision agent.

Challenges we ran into

The hardest part was making Qloo genuinely load-bearing rather than adding it as a decorative API call.

We designed the application so that Qloo entity IDs and Insights results directly determine the recommendation candidates.

Another challenge was avoiding overclaiming. Qloo affinities are aggregate cultural signals, not personal facts, so the interface explicitly shows limitations and confidence.

We also implemented explicit API failure handling instead of silently generating plausible-looking fallback recommendations.

Accomplishments that we're proud of

• Real Qloo entity resolution
• Real Qloo Insights queries
• Multi-stage agentic workflow
• Evidence critic and bounded retry path
• Qloo-versus-generic counterfactual
• Three judge-ready demo scenarios
• Public live deployment
• Public open-source repository
• 15 passing automated tests
• Server-side API key isolation

What we learned

The biggest lesson was that cultural intelligence changes the structure of an agent's decision process.

A generic model can describe what kind of place might suit someone.

Qloo gives the agent concrete cultural relationships to reason from.

That changes recommendation generation from:

"this sounds plausible"

to:

"these candidates emerged from the user's resolved cultural signals."

We also learned that cultural affinity should be treated as evidence, not certainty.

What's next for Pulse Taste Scout

The same pattern could expand into:

• travel planning
• hospitality concierge agents
• event discovery
• retail and brand discovery
• creator and audience matching
• culturally aware commerce agents
• taste intelligence for other autonomous agents

The long-term opportunity is to make cultural grounding a reusable intelligence layer for agents that need to make decisions involving human preferences.

Built With

• ai
• api
• css
• docker
• fastapi
• github
• html
• javascript
• pytest
• python
• qloo
• render