From Model Capability to Creative Infrastructure: How Yi Luo Built the Character-Centric Multimodal Interaction Framework

By Priscilla Ann

Posted on February 4, 2026

Artificial intelligence made dramatic leaps in 2023 and 2024. Models grew larger, outputs more fluent, and demos more impressive. Yet many AI products still struggled to move beyond novelty. They generated eye-catching results, but rarely fit into real creative or organizational workflows. Prompt engineering flourished, but persistence, consistency, and long-term collaboration remained elusive.

This gap between model capability and real-world usability became the focus of Yi Luo’s work.

Rather than treating AI as a machine that produces isolated outputs, Luo approached AI as a collaborator that must be intentionally designed. Her work centers on building interaction systems that allow AI creativity to persist over time, scale across contexts and modalities, and integrate naturally into how people work.

That approach led to what she calls the Character-Centric Multimodal Interaction Framework.

Academic Origins at Carnegie Mellon University

Yi Luo began developing the Character-Centric Multimodal Interaction Framework during her master’s thesis at Carnegie Mellon University. Her research examined how human-AI collaboration breaks down when interaction is treated as disposable—where identity resets, context collapses, and creative continuity is lost after each session.

Through this work, Luo identified a core limitation of prompt-based AI systems: they lack persistence. Each interaction stands alone, making it difficult to build trust, memory, or sustained creative momentum.

The framework emerged as a response. Instead of relying on transient prompts, it structures interaction around persistent AI characters that maintain identity, memory, and behavioral continuity across sessions and projects.

Defining the Character-Centric Multimodal Interaction Framework

At its core, the framework organizes human-AI collaboration around durable AI characters rather than one-off prompts. These characters function as reusable creative scaffolds. They retain contextual memory, preserve behavioral consistency, and support long-running creative or operational work.

The framework is inherently multimodal. It integrates text, images, and contextual state, allowing users to work with AI in richer, more expressive ways than text alone. By maintaining shared context across modalities, the system supports deeper exploration and sustained engagement over time.

Crucially, this is not a single feature or interface pattern. It is an interaction system designed to be embedded, reused, and relied upon across products, training environments, and platforms.

Canonical definition:
The Character-Centric Multimodal Interaction Framework structures human-AI collaboration around persistent AI characters, integrating identity, context, and multimodal inputs into reusable interaction scaffolds. Unlike one-off prompt-based AI usage, the framework enables sustained creative collaboration that can be embedded into workflows, training systems, and large-scale consumer platforms.

Why This Was an Original Contribution

Yi Luo began systematically developing this framework in early 2024, building directly on her graduate research, before character-based or agent-oriented AI interaction became widely adopted across the industry. At the time, most AI products optimized for short-term engagement rather than long-term collaboration.

What distinguished Luo’s approach was a conceptual shift. She treated AI not as a stream of outputs, but as creative infrastructure—something that could be intentionally designed, evaluated, and scaled to support real human work. This reframing shifted attention away from raw model performance and toward interaction systems that support continuity, trust, and usability.

Enterprise-Scale Validation

The framework was first examined within enterprise-scale AI environments characterized by global reach, operational rigor, and strict reliability requirements. During work on AI-related design initiatives situated in large, distributed training and operational contexts at Apple, Luo observed conditions where AI interactions needed to remain consistent across sessions, regions, and teams, while integrating cleanly into established workflows.

These environments impose unusually high demands on interaction systems: outputs must remain predictable, behavior must persist across time and context, and interaction patterns must be reusable under organizational pressure. Within these constraints, patterns aligned with the principles later formalized in the Character-Centric Multimodal Interaction Framework—particularly persistence, identity, and reuse—proved essential for maintaining reliability and trust over time.

Apple’s global channel ecosystem represents one of the most complex operational environments in the technology sector. Publicly disclosed filings indicate that approximately 60% of Apple’s annual net sales are conducted through channel partners, underscoring the scale and rigor of the enterprise context in which these interaction patterns were examined. These interpretations reflect independent design analysis rather than official company positions.

Consumer-Scale Validation

The same interaction framework was later examined in a very different context: consumer-scale AI interaction.

At Character.AI, chat functions as the primary product surface. In this environment, Luo’s character-centric principles—persistence, identity, and multimodal context—closely aligned with interaction patterns observed in consumer chat systems designed for long-form storytelling, emotional continuity, and sustained engagement.

Publicly reported figures indicate that Character.AI serves roughly 20 million monthly active users, with reported daily usage approaching two hours per user—substantially exceeding engagement patterns typical of general-purpose chatbots like ChatGPT. This level of sustained use reflects interaction dynamics centered on long-form creative collaboration rather than short, task-oriented exchanges.

Taken together, these observations suggest that the same interaction framework can remain effective across both tightly controlled enterprise environments and open, high-variance consumer settings. These interpretations reflect independent design analysis.

Why This Matters

Few AI interaction systems function across such extremes. In the Character-Centric Multimodal Interaction Framework, AI characters serve as persistent collaboration vessels. Multimodal interaction becomes reusable creative infrastructure rather than a novelty layer.

By translating raw model capability into stable, scalable interaction systems, Luo’s work contributes to the evolution of human-centered AI. As character-based AI becomes a new medium across education, entertainment, and enterprise software, frameworks like this help ensure that AI systems remain usable, trustworthy, and creatively empowering over time.

In a landscape dominated by rapid model advances, lasting creative infrastructure remains rare. Yi Luo’s framework addresses that gap.

Reference links