Case Study - Summarized Version

Case Study - moderately detailed

Project Overview

This dual-platform ecosystem addresses museum accessibility and visitor churn by translating raw LLM logic into intuitive user experiences. The platform utilises a deterministic-to-probabilistic UI bridge, mapping dual interactive sliders (Tone and Recommendations) directly to backend LLM hyperparameters. To ensure an inclusive, distraction-free environment for marginalized visitors, the next phase introduces hidden prompt-channeling layers: routing inputs across 5 psychology-trained AI personas while strictly regulating token output lengths to prevent mobile viewport text walls.

Key Technical Pillars to highlight lower down:

  • Dual-State Sliders: Translating high-level user preferences into real-time context injections.

  • Contextual Agent Routing: Designing the behind-the-scenes system architecture that matches user psychologies to a specific museum expert persona.

  • Token Containment & Truncation: Setting hard limits on AI output to avoid user cognitive fatigue on mobile screens.

Psychology-driven UX architecture: 5 distinct agent profiles (how inputs are routed contextually) under the hood:

  • The Empathetic Storyteller (Psychology Focus: Belonging & Low Confidence)

    • Target User: Visitors who feel intimidated, out of place, or alienated by traditional art history.

    • System Prompt Boundary: Use warm, conversational language. Avoid jargon. Frame art through human emotions, universal themes, and personal histories rather than dates and technical eras.

  • The Curiosity Curator (Psychology Focus: Cognitive Engagement & High Autonomy)

    • Target User: Visitors who want to explore hidden details and love solving puzzles or uncovering secrets.

    • System Prompt Boundary: Frame descriptions around unanswered questions, historical mysteries, and eccentric creator details. End responses with open-ended prompts to guide spatial exploration.

  • The Sensory Anchor (Psychology Focus: Neurodivergence & Sensory Processing)

    • Target User: Visitors with ADHD, autism, or sensory sensitivities who need a calming, highly structured environment.

    • System Prompt Boundary: Keep descriptions ultra-concise with bullet points. Focus on tactile, spatial, and auditory elements of the artwork. Avoid sensory-overloading metaphors and include low-traffic spatial navigation tips.

  • The Socio-Cultural Analyst (Psychology Focus: Critical Thinking & Diverse Identities)

    • Target User: Marginalized or activist visitors looking for representation, equity, and the true political context of art.

    • System Prompt Boundary: Analyze pieces through a lens of power dynamics, colonialism, gender, and class struggle. Re-center marginalized voices that standard museum plaques often omit.

  • The Playful Synthesizer (Psychology Focus: Low Attention Span & High Fatigue)

    • Target User: Tired visitors, younger audiences, or those experiencing "museum fatigue" who want quick, high-energy connection.

    • System Prompt Boundary: Use highly accessible analogies, modern pop-culture comparisons, and touchpoints of humor. Keep the token length strictly short to ensure rapid reading.

Behind the scenes — Changes to MUSE (AI Guide) for better capturing words and their impact

  • Goal: Make MUSE count and weigh words more accurately to give clearer guidance.

    Planned changes

    • Group multi-word units (names, idioms, compounds) as single units.

    • Give words dynamic weights based on role in the sentence, importance, and context; normalize scores per input.

    • Extend context for key phrases, drop low-relevance content, and focus on high-weight phrases.

    • Use phrase-level embeddings for clearer combined meaning.

    • Run a quick sense check to pick likely meanings before weighting.

    Expected results

    • Keep phrase meanings intact.

    • Align guidance better with user intent.

    • Reduce hallucinations by focusing on high-impact words.

    • Improve UX by highlighting important phrases.

    Next steps

    • Run A/B tests and adjust thresholds.

    • Grow the annotated corpus across domains and languages.

    • Test UI options to show phrase importance and collect feedback.

The full Responsive Museum Platform case study (including detailed steps and interactive prototypes) is available on Behance

Previous
Previous

Free me! - Vegan Live Page (AI-Native & Code-Based Prototyping)

Next
Next

Live-Music Network (Complex Multi-Role Ecosystems)