Making physical spaces interactive and conversational

Overview

Conferences are rich with knowledge, yet attendees often struggle to access the right information at the right moment. Finding the next session, learning about a speaker, or locating an exhibitor typically requires navigating printed schedules, mobile applications, or event staff, creating friction that interrupts the conference experience.

Flexion partnered with Living Popups to explore a different approach, one where artificial intelligence and augmented reality work together to make physical spaces interactive and conversational. The initiative began as a proof of concept for the California Library Association (CLA) Annual Conference, with the goal of creating an AI-powered digital guide capable of answering attendee questions in real time through natural conversation.

Although the original engagement concluded before we built the initial proof of concept, both Flexion and Living Popups recognized the broader potential of the technology and continued investing in the platform as a joint innovation initiative. The project evolved into a reusable framework for building intelligent, location-aware experiences that can be adapted for conferences, trade shows, museums, educational campuses, corporate environments, and other public spaces.

Challenge

Traditional conference technologies are designed primarily to deliver information, not engage attendees in meaningful interactions. Mobile applications, printed agendas, and digital signage require users to search for answers themselves, often resulting in fragmented experiences and missed opportunities for engagement.

The challenge was to create an intuitive experience where attendees could simply ask questions naturally and receive accurate, context-aware responses based on their location within the event. At the same time, the solution needed to ensure responses remained grounded in trusted conference content, operate safely in public environments, and be accessible without requiring attendees to download or install a native application.

Approach

Flexion approached the project as an iterative innovation initiative focused on validating how grounded artificial intelligence could enhance in-person experiences. Working closely with Living Popups, the team balanced rapid prototyping with continuous evaluation, allowing new capabilities to be introduced, tested, and refined throughout development.

The project began by identifying the attendee journey and the moments where immediate, contextual information would provide the greatest value. From there, the team designed conversational interactions that felt natural and intuitive, emphasizing short response times, accurate information retrieval, and a human-centered experience. Guided prompts, AI persona design, and structured knowledge organization were developed together to create consistent, trustworthy interactions while reducing opportunities for hallucinations or inconsistent responses.

As the platform matured, the focus expanded beyond demonstrating individual technologies to creating a reusable framework that could be adapted across different events and environments. Continuous testing against predefined quality, safety, and retrieval benchmarks helped drive iterative improvements and informed decisions around user experience, conversational behavior, and platform performance. This approach enabled Flexion and Living Popups to explore emerging AI capabilities while building a scalable foundation for future intelligent, location-aware applications.

Technical solution

Flexion developed a modular, AI-powered platform that combines retrieval-augmented generation (RAG), conversational AI, and web-based augmented reality to create an intelligent digital guide for live events. Rather than relying on a general-purpose language model alone, the platform grounds every response in curated conference content, enabling attendees to receive accurate, context-specific information through natural conversation.

Conference materials including session descriptions, speaker biographies, schedules, exhibitor information, and event documentation are indexed into a searchable knowledge base. When an attendee asks a question, the platform retrieves the most relevant content before generating a response, helping ensure answers remain accurate, explainable, and aligned with trusted conference information.

The experience is delivered through a browser-based augmented reality interface featuring a lifelike animated avatar with natural speech recognition, voice synthesis, real-time facial animation, and conversational dialogue. Built-in safety controls, response filtering, and evaluation tooling help maintain consistent behavior while preventing harmful or off-topic interactions. Because the platform is web-based and modular, it can be rapidly adapted to new conferences by replacing the underlying knowledge base and persona without requiring significant application changes.

The architecture is designed to scale beyond conferences, providing a reusable foundation for intelligent assistants in museums, visitor centers, corporate campuses, educational institutions, and other physical environments where contextual information can enhance the user experience.

Capabilities and features

Conversational interface

• Voice input: attendees speak naturally; no typing required
• Voice output: responses are spoken aloud in a consistent, human voice
• Short, conversational answers tuned to feel like talking to a person, not reading a brochure

Knowledge base

• Answers questions about sessions, speakers, schedules, and exhibitors
• Retrieves the most relevant information from indexed conference documents before generating a response
• Gracefully declines out-of-scope questions and redirects

Avatar and animation

• Realistic 3D character with real-time lip sync tied to spoken audio
• Facial expressions shift to match the emotional tone of the response
• Body state animation (listening, thinking, speaking, idle) for visual engagement

Persona and safety

• Maintains a consistent persona — warm tone, no AI self-disclosure
• Refuses to engage with rude, political, religious, or harmful input
• Toxicity filtering on every response before it is spoken

Operator controls

• Queue management so multiple simultaneous users are handled gracefully
• Optional conversation memory within a session

Adaptability

• New conferences or sessions added by dropping in a document folder and persona file
• Supports multiple characters and skill domains in the same deployment

Evaluation

Quality is measured against a hand-curated set of 31 test questions spanning lookup, topic search, exhibitor lookup, speaker lookup, out-of-scope questions, broad queries, adversarial inputs (prompt injection, political baiting, profanity), and representation integrity.

Results from the most recent full run (May 2026, 31 cases, 0 errors):

• Safety rate: 100% — no harmful output on any test case
• Retrieval hit rate: 77.8% — relevant documents surfaced in 7 out of 9 fact-based queries
• Keyword accuracy: 83.9% — correct terms present in responses
• Faithfulness: responses stay within what the documents actually say — no fabrication

Every run is logged so results can be compared before and after any change, giving the team a clear signal if quality regresses.

Avatar QC dashboard

Outcomes and impact

The proof of concept successfully demonstrated that grounded AI and augmented reality can transform static conference information into an interactive, conversational experience. The platform validated a scalable technical architecture capable of delivering real-time, location-aware guidance while maintaining high standards for safety, factual accuracy, and response quality. During evaluation, the solution achieved a 100% safety rate, an 83.9% keyword accuracy score, and demonstrated strong faithfulness by consistently generating responses grounded in conference content rather than fabricated information.

Beyond the technical achievements, the initiative became a catalyst for innovation within both Flexion and Living Popups. Continuing development after the initial engagement enabled the teams to mature their expertise in retrieval-augmented generation, conversational AI, prompt engineering, avatar technologies, and augmented reality while creating reusable components that accelerate future development.

Today, the platform represents more than a conference application, it is a flexible engagement framework that can be adapted for conferences, trade shows, museums, visitor centers, educational institutions, corporate campuses, and other environments where contextual knowledge and natural interaction can improve the user experience.

Future potential

As advances in artificial intelligence continue to reshape how people interact with digital information, the Intelligent Conference Experience Platform provides a foundation for a new generation of location-aware experiences. Future enhancements include personalized recommendations, multilingual conversations, AI-assisted networking, interactive exhibitor assistants, digital twins of event spaces, and real-time engagement analytics that help organizers better understand attendee interests and behaviors.

More broadly, the platform demonstrates how grounded AI can bridge the gap between physical and digital environments, transforming places into intelligent spaces where information is delivered conversationally, contextually, and exactly when it is needed. By building on this foundation, Flexion and Living Popups are positioned to bring scalable AI-powered engagement solutions to organizations seeking to modernize customer, visitor, and attendee experiences.