Inspiration
Creator discovery is usually a mix of keyword search, follower counts, and intuition. That can find popular creators, but it struggles with a more important question: will this creator feel culturally relevant to a specific campaign in a specific market?
We wanted to build something that goes beyond "find fashion creators in Mumbai." CultureMatch asks: which creators' observed content actually resonates with the cultural signals behind this fashion campaign in Mumbai?
Qloo made that possible. Its Taste Graph gives the system cultural grounding that an LLM or a conventional influencer-search tool cannot reliably infer on its own.
What it does
CultureMatch is an agentic creator-discovery platform for regional marketing campaigns.
A marketer submits a campaign brief with product context, target market, cultural references, budget, platform, and audience constraints. CultureMatch then:
- Resolves campaign concepts, references, and tags with Qloo.
- Builds a market-aware cultural fingerprint using Qloo signals.
- Discovers creator candidates through Apify.
- Resolves observed creator content into cultural signals.
- Uses Qloo to measure campaign-to-market-to-creator-content resonance.
- Ranks creators with transparent, deterministic scoring.
- Shows evidence, caveats, missing data, and how ranking changes with Qloo cultural affinity removed.
The result is an explainable shortlist, not just a list of creators with matching keywords.
How we built it
We built CultureMatch with Next.js, TypeScript, Supabase, Qloo, Apify, OpenRouter, and the Vercel AI SDK.
Qloo is the core intelligence layer:
- Campaign language and cultural references are resolved into Qloo entities and tags.
- Qloo regional signals build a cultural fingerprint for the target market.
- Creator bio and observed content signals are resolved through Qloo.
- Creator signals are matched against campaign signals in the selected market.
- Cultural affinity contributes to the final creator-fit score only when evidence is available.
Apify provides broad creator discovery. Supabase stores campaigns, workflow state, provider results, evidence, and scored recommendations with owner-scoped access controls.
We use bounded agents for research planning, evidence assessment, and optional search refinement. Their outputs are schema-validated, tool calls are budgeted, and deterministic fallbacks keep the workflow functional when a model is unavailable. The final ranking remains deterministic and evidence-backed.
Challenges we ran into
The hardest challenge was avoiding superficial Qloo integration.
Early versions could identify cultural signals in a market, but that alone did not prove a creator matched a campaign. A creator could score highly simply because their topic was popular locally. We redesigned the scoring flow so creator-content signals must connect to campaign signals and the target market.
Other challenges included:
- Resolving ambiguous campaign descriptions without converting every word into an unreliable cultural tag.
- Keeping creator evidence honest when provider data is incomplete.
- Handling asynchronous Apify runs, webhooks, retries, and idempotency.
- Preventing agents from creating excessive provider calls or uncontrolled refinement loops.
- Explaining uncertainty clearly: creator location is not audience geography, and cultural affinity is not guaranteed ROI.
Accomplishments that we're proud of
- Built a genuine campaign x market x creator-content cultural matching path with Qloo.
- Made Qloo's impact inspectable through a with-Qloo versus without-Qloo ranking comparison.
- Kept scoring transparent: each recommendation includes component scores, evidence, availability states, and caveats.
- Built agents that are useful but bounded, rather than allowing an LLM to make opaque final decisions.
- Preserved unavailable provider data instead of inventing fees, audience geography, or performance signals.
- Added regional cultural exploration through Qloo comparisons, heatmaps, and trending public figures.
What we learned
We learned that cultural intelligence is most useful when it changes a decision, not when it is displayed as an extra data point.
Qloo became meaningful only after it was placed at the center of the matching model. The important relationship is not "what is popular in this market?" It is "which creator's demonstrated cultural signals resonate with this specific campaign in this market?"
We also learned that agentic workflows need strong boundaries. Agents are valuable for deciding what to investigate next, but structured tools, explicit budgets, schema validation, deterministic fallbacks, and evidence-backed scoring are essential for reliability.
What's next for Culture Match
Next, we want to:
- Support more creator platforms and richer post-level content evidence.
- Improve market and audience conditioning with verified provider data.
- Add campaign performance feedback loops to evaluate whether cultural resonance predicts better outcomes.
- Let teams compare multiple markets before launching a campaign.
- Add collaborative shortlists, approvals, and campaign reporting.
- Expand Qloo-powered cultural exploration into creative strategy, content themes, and regional launch recommendations.
Our goal is to help marketers move from "who has reach?" to "who already speaks the cultural language of this campaign?"
Built With
- ai-sdk
- next.js
- openrouter
- qloo
- vercel
