Asterisk Taste: Sovereign Qloo Cultural Agent & Bridge

Cultural intelligence agent grounded in Qloo's 250M+ graph, formally verified in SPARK Gold, Latin semantic interlingua, and accessible 3D sign avatar.

Comment

Asterisk Taste: Sovereign Cultural Intelligence Agent & Latin Semantic Bridge

Hackathon: Qloo Agentic Hackathon ($25,000 Prize Pool)
Live Endpoint: https://web4light.online/
Source Repository: https://gitlab.com/web4light/Universal-Translation-Layer-app
License: MIT License
Formal Verification: Ada/SPARK 2022 Level 4 Gold (DO-178C DAL-A)

🌟 Inspiration & The Problem: Beyond „Cultural Beige“

Mainstream Large Language Models are culturally flat. When asked for recommendations, they gravitate towards generic, predictable averages — the dreaded „cultural beige“. Worse yet, typical recommendation systems operate as privacy black holes, vacuuming up sensitive user telemetry into centralized advertising databases.

Asterisk Taste solves this by uniting Qloo’s Taste Intelligence Graph (250M+ entities) with a mathematically proven, privacy-first sovereign architecture. Instead of guessing what people enjoy, Asterisk Taste grounds every agent reasoning step in real cultural affinity data across dining, music, film, literature, fashion, and architecture, while ensuring that the user’s personal data is protected by formal verification and sovereign cryptography.

🏛️ Key Architectural Innovations

1. Formally Verified Cultural Guards (Ada/SPARK 2022 Level 4 Gold)

While standard agent frameworks rely on loose prompt engineering, Asterisk Taste enforces mathematical invariants on taste lookups:

  • Cultural_Taste_Semantic_Guard (7 VCs): Guarantees that every taste affinity lookup is bounded, type-safe, and grounded in validated Qloo entity vectors. Zero out-of-bounds hallucinations.
  • Cross_Domain_Taste_Synergy_Governor (8 VCs): Performs cross-domain taste synthesis (e.g., mapping culinary preferences to architectural styles or cinematic tastes to musical acoustic profiles) using differential privacy ($\epsilon$-DP) guarantees.

2. Latina jako sémantická mezivrstva (Classical Latin Semantic Interlingua)

To preserve sovereign privacy, raw user prompts are never transmitted in clear text across the wire. Asterisk Taste translates concepts into Classical Latin semantic lemmas (musica, cinemateca, gastronomia, architectura, vestis). The external network only observes cryptographically hashed Latin tokens and resonant frequencies; user identities remain strictly airgapped.

3. Bezbariérová rampa (Radical Accessibility)

Cultural taste belongs to everyone:

  • Deaf Accessibility: Real-time 3D sign language avatar embedded directly in WebXR / WebGL (web4light.online), translating cultural insights into expressive spatial signing.
  • Blind Accessibility: Sovereign Voice Vault utilizing natural expiratory air stream phonics (true speech on the natural breath stream, without synthetic gasps).

4. Nula je nebinární nula (Mathematical Privacy at 04:23 UTC)

At 04:23 UTC daily, the in-memory RAM scrub executes an unconditional purge of volatile query memory. Zero data retention means user preferences cannot be leaked, subpoenaed, or monetized.

🔄 End-to-End Workflow & Qloo Integration

  1. User Inquiry: The user requests recommendations (e.g. „Curate a weekend journey for an architect who loves post-rock and minimalist Japanese dining“).
  2. Sovereign Interlingua Extraction: The query is decomposed into canonical affinity vectors via Latin semantic tokens.
  3. Qloo API Taste Graph Dispatch: Asterisk Taste queries the Qloo API across multiple domains:
    • Dining: High-affinity dining entities matched to Japanese minimalism.
    • Music & Audio: Ambient and post-rock acoustic profiles.
    • Architecture & Places: Spatial destinations exhibiting high cross-domain resonance with the user's aesthetic taste.
  4. SPARK Verification Gate: Cross_Domain_Taste_Synergy_Governor checks the affinity matrix invariants before synthesizing the itinerary.
  5. Multi-Modal Rendering: The result is delivered simultaneously in structured markdown, natural breath-flow audio, and WebXR 3D sign avatar.

🛠️ Verification & Testing Instructions

1. Live Interactive Web Experience

Visit https://web4light.online/ to interact with the live spatial frontend, test the cultural recommendation flows, and experience the 3D sign avatar in browser.

2. Local Source & Proof Verification

# Clone the public repository
git clone https://gitlab.com/web4light/Universal-Translation-Layer-app.git
cd Universal-Translation-Layer-app

# Inspect Qloo integration modules:
# - Cultural_Taste_Semantic_Guard (spark/src/cultural_taste_semantic_guard.ads)
# - Cross_Domain_Taste_Synergy_Governor (spark/src/cross_domain_taste_synergy_governor.ads)

# Run Level 4 Gold formal proof:
gnatprove -P default.gpr --level=4 --report=all

All verification conditions prove with 0 errors, 0 warnings, 100% proof coverage.

đź“ś Open Source & Community

Released under the MIT License. Fully open-source, mathematically proven, and grounded in real cultural intelligence.

Built With

Share this project:

Updates

Submission history