AURA (CultureGuard)

Autonomous cultural red-teaming and brand fit auditor grounded in Qloo's 250M+ behavioral taste graph.

Comment

Inspiration

Every year, enterprise brands and sponsorship agencies waste tens of millions of dollars on celebrity endorsements, sleeve deals, and festival sponsorships that trigger immediate consumer backlash or zero sales conversion.

Standard LLMs (ChatGPT, Claude, Gemini) consistently fail at brand due diligence because of the "Synergy Sycophant" trap: when asked whether Chelsea FC should sign an eight-figure sleeve deal with Prime Hydration, or whether Target should sell luxury designer evening gowns, ungrounded AI hallucinates approval:

"This is an exciting and modern partnership! Both brands are globally renowned with passionate followings. The contrast creates incredible viral buzz!"

Standard LLMs operate on textual co-occurrence and polite marketing optimism. They have zero access to actual consumer transactions, dining habits, or cultural affinities. What they cannot see are the measurable behavioral cliffs:

  • Demographic Chasm: A £1,500 West-Stand season ticket holder in London has near-zero lifestyle overlap with an adolescent influencer energy drink.
  • Aesthetic Dissonance: Rave subcultures perceive corporate Wall Street wealth management as an intrusive institutional invasion.
  • Capital Dissipation: Millions in marketing spend get destroyed on deals that generate brief Twitter memes followed by customer alienation.

We built AURA to give agents what they’ve always lacked: objective cultural taste telemetry grounded in Qloo’s 250M+ entity graph.

What it does

AURA is an autonomous, tool-augmented brand auditor that stress-tests proposed commercial partnerships before contracts are signed and capital is deployed.

  1. Adversarial Brand Fit Scoring: Evaluates proposed partnerships across three weighted dimensions using mathematical models: $$\text{Composite Fit Score} = 0.35 \times \text{Demographic} + 0.40 \times \text{Cultural} + 0.25 \times \text{Aesthetic}$$
  2. Illustrative Capital Exposure Modeling: Quantifies financial downside before committing capital: $$\text{Capital at Risk} = \text{Deal Value} \times \left(1 - \frac{\text{Score}}{100}\right)$$
  3. Competitor & Exclusivity Detection: Catches the dangerous High-Affinity Cannibalization Trap. When direct rivals (e.g., Coca-Cola × Pepsi or Nike × Adidas) are evaluated, AURA recognizes that high raw consumer overlap is actually a zero-sum commercial collision that breaches standard category exclusivity covenants.
  4. Autonomous Graph Repair: When an evaluation reveals high friction ($< 40/100$), AURA traverses nearest-neighbor clusters in Qloo’s graph to surface high-converting alternative partners that bridge the divide.
  5. Closed-Loop Counterfactual Verification: Re-scores replacement partners using independent tag concordance from Qloo’s /v2/tags to verify quantitative score lift ($+60$ to $+80$ pts) before compiling an investment-grade Executive Due Diligence Memo.

How we built it

AURA is engineered as a hybrid agentic architecture pairing deterministic policy routing with live Qloo tool execution and semantic synthesis:

  • Qloo Hackathon Sandbox Integration (https://hackathon.api.qloo.com):
    • tool_qloo_entity_resolution: Queries /search?query=... to resolve entities into canonical URNs and taxonomy classifications.
    • tool_qloo_affinity_telemetry: Queries /v2/insights?filter.type=...&signal.interests.entities=... to pull high-dimensional co-preference matrices across music, dining, fashion, and lifestyle.
    • tool_qloo_tag_discovery: Queries /v2/tags?filter.query=... to evaluate independent aesthetic property and lifestyle Jaccard concordance.
  • Autonomous 5-Step ReAct Loop: Features real thought-action-observation telemetry, execution latency tracking, and transparent data source pills ([Live Qloo API • 200 OK • 140ms] vs [Archived Telemetry Snapshot]).
  • Frontend Craftsmanship: Built with React 19, TypeScript, and Vite. We engineered a custom Apple-inspired dark-mode glassmorphic design system (#000000/#161617, Apple Blue #2997ff) with an interactive SVG force-directed cultural network graph.

Challenges we ran into

Our most critical architectural breakthrough happened when stress-testing direct competitors like Coca-Cola × Pepsi:

In pure vector space, Coca-Cola and Pepsi have a 95% taste correlation because their consumers eat at the same restaurants, watch the same sports, and listen to the same pop music. A naive recommendation engine happily recommended them as an ideal brand partnership!

In the real commercial world, partnering with your direct substitute is commercial suicide: it breaches universal sponsorship Category Exclusivity Clauses, triggers consumer brand confusion, and produces zero net-new audience discovery.

We solved this by engineering an Adversarial Category Exclusivity Detector. The agent recognizes when two entities belong to the same tier-1 commercial cluster, flags the High-Affinity Cannibalization Paradox, applies an exclusivity penalty (dropping the composite score from 95 to 14/100), and autonomously routes the repair loop toward high-affinity cross-domain partners (such as Formula 1, Spotify, or Riot Games) that deliver actual incremental reach.

Accomplishments that we're proud of

  • Zero-Hallucination Grounding: Unlike conversational bots that invent opinions, AURA’s scorecard, friction points, and replacement partners are driven by Qloo's empirical data.
  • Closed-Loop Self-Correction: When a deal fails, AURA doesn't just critique—it self-corrects by surfacing and validating replacement candidates with measurable score deltas.
  • Enterprise-Ready Due Diligence Memos: Generates an exportable, C-suite-ready memo complete with capital risk estimates, mathematical breakdowns, and strategic directives.
  • Sub-150ms Performance: Instantaneous local and client-side processing paired with optimized API telemetry.

What we learned

  • High vector similarity does not equal strategic commercial viability. An auditor must account for real-world market dynamics (cannibalization, exclusivity covenants, brand prestige).
  • True agentic utility lies in critique-and-repair. Pointing out problems is easy; autonomously traversing a 250M+ node graph to find validated, high-converting alternatives is where real business value is created.

What's next for AURA (CultureGuard)

We are positioning AURA as an automated B2B intelligence platform for sponsorship valuation agencies (CAA, WME, Omnicom), corporate brand marketing teams, and private equity due diligence firms. Future milestones include:

  • Multi-brand portfolio optimization (balancing an entire kit or festival roster to maximize unique audience reach without duplicate overlap).
  • Automated contractual term-sheet risk generation (generating contract clauses based on identified friction dimensions).
  • Enterprise CRM & Pitch Deck integrations.

Built With

  • artificial-intelligence
  • autonomous-agents
  • brand-safety
  • cultural-intelligence
  • data-visualization
  • due-diligence
  • financial-modeling
  • machine-learning
  • qloo
  • qloo-api
  • react
  • react-19
  • react-agent
  • svg
  • tailwindcss
  • taste-ai
  • typescript
  • vercel
  • vite
Share this project:

Updates

Submission history