Qloo Agentic Hackathon / 2026

Find the markets that already share your taste.

TasteBound builds a Qloo taste profile for a destination, scores origin markets by taste lift over a popularity baseline, compares that with real arrivals — and shows you the untapped markets worth a campaign.

Taste fit, not a visitor forecastStored sample, no sign-up

QLOOTASTE SIGNAL / 01TASTE FITNOT A FORECASTTB / 26UNTAPPED / 01ARRIVALS
Physical signals / digital tasteQloo entity graph
Method / three moves

From a destination to a decision.

No black boxes, no visitor claims. Each step is deterministic code over Qloo responses, with the model limited to proposals and prose.

Step 01
Profile

Describe the destination

Name it, add signature dishes, landmarks, music and more. TasteBound resolves each item to a real Qloo entity and shows what it could not resolve.

Step 02
Score

Measure taste lift

Every origin market gets a taste score from Qloo affinity lift over a popularity baseline, with coverage, confidence and thin-data flags in the open.

Step 03
Compare

Find the gap

Bring your arrivals CSV. The gap quadrant shows where taste is strong but arrivals are not — with audiences, angles, partners and a budget split.

The gap quadrant

Strong taste. Few arrivals.

Taste score on one axis, arrivals share on the other. Untapped markets sit top-left: strong taste fit, few visitors today. Every point opens the full market brief.

Audit trail

Every number has a source.

Each score, gap, audience and partner links back to the Qloo call that produced it. The trace page lists every call, tool step and warning in order.

Taste score
percentile of lift
Lift
target − baseline
Gap
taste − arrivals
Open the sample run
Principles / non-negotiable

Qloo decides. The model narrates.

The agent proposes item names, chooses tools and writes briefs that must pass a grounding validator. Scores, gaps, audiences, angle evidence and partners stay in deterministic code.

01 / Signals

Scores from Qloo

Affinity lift over a popularity baseline. Percentile-ranked within the run, never dressed up as a visitor forecast.

02 / Language

Briefs with receipts

Model-written prose is labelled and checked against evidence. Without a model, the template is labelled too.

03 / Failure

Fail closed

No Qloo key, no ranking. The kill switch is a tested feature, not an error path that quietly invents numbers.

Built with

Taste graph in. Campaign plan out.

One portable stack: Qloo for taste, your data for arrivals, a grounded agent for the words.

Qloo Taste GraphYour arrivals CSVGemma 4 on Ollama CloudNext.js + FastAPI
Start with one destination

Find the markets already leaning in.

The sample is a stored, read-only run for a real destination — no Qloo quota, no sign-up. Or bring your own destination and arrivals.

Taste fit, not a visitor forecast. No visitor claims, no black boxes.