Qloo Cultural Taste Agent

AI Agent Framework with Cross-Domain Cultural Intelligence & Qloo Taste Graph Grounding

Comment

Qloo Cultural Taste Agent 🍸🎭

Submitted to the Qloo Agentic Hackathon ($25,000 Cash Prizes + $25,000 Investment Bonus by Jason Calacanis)
Hosted by: Qloo on Devpost
Judges: Jason Calacanis, Mike Diolosa (CTO, Qloo), Nicole Seligman (OpenAI Board of Directors), Todd Boehly (CEO, Eldridge), Cedric the Entertainer
Live Interactive Studio: https://liutingqiu.github.io/qloo-cultural-taste-agent/
GitHub Repository: https://github.com/liutingqiu/qloo-cultural-taste-agent
Author & Lead Engineer: Tingqiu Liu (@liutingqiu, [email protected])
Test Status: 6/6 Automated Unit Tests Passing (100% Pass Rate)

🌟 Executive Summary: Cultural Grounding for Autonomous AI Agents

Modern AI agents (Claude, AutoGen, CrewAI) excel at logical reasoning, multi-step tool execution, and code generation. However, they are culturally blind. When asked to recommend a restaurant for an indie music lover, design an aesthetic itinerary, or pair wine with music, generic LLMs fall back on generic cliches and popular averages:

  • They guess based on token co-occurrence rather than multidimensional taste resonance.
  • They cannot differentiate between commercial popularity and genuine subcultural coherence (e.g. pairing aggressive fast food with meditative ambient vinyl listening).
  • They lack an objective metric for Cultural Coherence.

Qloo Cultural Taste Agent equips autonomous AI agents with Qloo's cultural intelligence (250M+ entities across Music, Dining, Film, Fashion, and Travel). By mapping high-dimensional aesthetic vectors, energy profiles, exclusivity indices, and cultural eras, this agent produces recommendations and itineraries that are genuinely culturally grounded.

🏗️ System Architecture & Model Context Protocol (MCP)

                  [User Natural Language Intent]
                                │
                                ▼
         [CulturalAgentPipeline: Planner & Entity Resolver]
                                │
                                ▼
         ┌──────────────────────────────────────────────┐
         │       Model Context Protocol (MCP) Server     │
         │                                              │
         │  1. qloo_resolve_entity                      │
         │  2. qloo_compute_taste_affinity              │
         │  3. qloo_cross_domain_synthesis              │
         │  4. qloo_audit_cultural_fit                  │
         └──────────────────────┬───────────────────────┘
                                │
                                ▼
                 [Qloo Taste Graph Core Engine]
  ┌──────────────┬──────────────┬──────────────┬──────────────┐
  │    MUSIC     │    DINING    │     FILM     │   FASHION    │
  │ Shoegaze     │ Natural Wine │ Neo-Noir     │ Issey Miyake │
  │ Modal Jazz   │ Kissaten     │ Cyberpunk    │ Workwear     │
  │ Ambient/IDM  │ Fermentation │ Arthouse     │ Minimalist   │
  └──────────────┴──────────────┴──────────────┴──────────────┘
                                │
                                ▼
       [Cultural Coherence Index (CCI) & Dissonance Audit]
                                │
                                ▼
        [Culturally Grounded Itinerary & Reasoned Output]

⚡ Key Capabilities & Technical Features

1. Cross-Domain Cultural Vector Modeling

Each entity in the Qloo Taste Graph is characterized across 4 continuous vector dimensions:

  • Aesthetic Core: e.g., ethereal-distortion, minimalist-sophistication, warm-artisanal, futuristic-glitch, brutalist-clean.
  • Energy Level (0.00 – 1.00): Ambient tranquility vs kinetic high-tempo intensity.
  • Exclusivity Index (0.00 – 1.00): Mainstream universality vs niche subcultural specialization.
  • Cultural Era: Cross-generational context from late-50s-timeless to 90s-indie and contemporary-nordic.

2. Standardized Model Context Protocol (MCP) Tools

Enables any autonomous LLM agent (Claude Desktop, Cursor, OpenAI, LangChain) to interface natively with Qloo:

  • qloo_resolve_entity: Disambiguates search keywords into canonical Qloo taste nodes.
  • qloo_compute_taste_affinity: Calculates calibrated mathematical affinity (0.00 to 1.00) between any two entities across different domains.
  • qloo_cross_domain_synthesis: Given anchor tastes, synthesizes culturally coherent items across desired target domains.
  • qloo_audit_cultural_fit: Audits multi-entity bundles and itineraries to detect aesthetic dissonance before presentation.

3. Interactive Live Workbench

Hosted on GitHub Pages with:

  • Dynamic Taste Network Graph: Visual SVG relationship map illustrating active cross-domain affinity lines.
  • One-Click Curated Presets: Instant demonstration of shoegaze/natural wine, modal jazz/kissaten, and IDM/cyberpunk synthesis.
  • Real-Time MCP Execution Trace: Transparent inspectability of every agent tool call, arguments, and observations.

🧪 Verification & Test Suite

The project includes 6 automated unit tests with zero external test framework dependencies:

cd qloo-cultural-taste-agent
npm test

Test Results:

✔ 1. QlooClient resolves entities by ID, genre keyword, and filters by domain
✔ 2. computeTasteAffinity calculates calibrated cross-domain resonance scores
✔ 3. TasteGraphEngine synthesizes culturally grounded multi-domain recommendations
✔ 4. auditCulturalCoherence correctly evaluates cultural harmony and detects aesthetic dissonance
✔ 5. QlooMcpServer provides valid MCP tool specifications and executes tool calls
✔ 6. CulturalAgentPipeline end-to-end planAndExecute produces verified reasoning trace and itinerary
ℹ tests 6
ℹ pass 6
ℹ fail 0
ℹ duration_ms 97ms

📄 Open Source License

Licensed under the MIT License.

Built With

Share this project:

Updates

Submission history