Pushing Practical AI at SIGGRAPH 2025: Digital Humans in the Real World
In this blog
At SIGGRAPH 2025, World Wide Technology (WWT) took the stage—not with a booth, but with bold ideas. In a 30-minute thought-leadership session led by Kyle Lindsey, Director of Innovation at WWT, attendees were immersed in the trials, tribulations, and triumphs of building digital humans for real-world enterprise use.
From lab to live: The challenge of real deployment
Kyle's session, titled "Digital Humans in the Real World," pulled back the curtain on what it takes to bring digital humans out of the lab and into production. It's not just about flashy avatars or clever chatbots—it's about shipping, receiving, energy, and connectivity. These aren't just technical hurdles; they're the reality of deploying AI at scale in large enterprises.
WWT's journey began with RAGbot, a retrieval-augmented generation system that fused LLMs with private enterprise data. This evolved into Ellie, a multilingual digital human powered by NVIDIA ACE stack, NVIDIA® Riva ASR/TTS, Audio2Face and rendering with NVIDIA RTX™. Ellie wasn't just a demo—she was a fully offline, air-gapped kiosk running Llama-3 8B across five languages on a quad NVIDIA RTX 6000 Ada Generation server.
The full AI stack: Why it matters
Kyle emphasized that building digital humans is hard—and what makes it hard is the sheer number of moving parts. From speech recognition and translation to rendering and compliance, every layer of the AI stack must work in harmony. WWT's tech stack included:
- NVIDIA Riva for low-latency speech recognition and natural-sounding TTS
- NVIDIA NeMo™ Retriever for grounding LLM responses in enterprise knowledge
- Audio2Face for real-time facial animation
- RTX for lifelike rendering and collaboration
This wasn't just a showcase of tools—it was a demonstration of orchestration. The pipeline had to convert speech to text, reason over enterprise data and render a response in real time. And it had to do all of this in noisy, unpredictable environments like conferences and hospitals.
ElevenLabs and the next generation: Meet Kendra
While Ellie was the star of the show, Kyle offered a glimpse into the future with Kendra, WWT's next-gen digital human. Kendra upgrades Ellie's voice with ElevenLabs technology, boosts realism with Unreal Engine 5.6 Metahuman, and runs on GPT-4o for smarter, more expressive interactions.
Memory optimization, as well as AI and Visual rendering pipelines, is still evolving, along with system availability. As Kyle put it, "This stuff is hard"—and that's exactly why it matters.
Real-world scenarios: From hospitals to movie theaters
Kyle's presentation didn't just focus on tech—it explored how digital humans can transform everyday experiences:
- Movie theaters with loyalty-aware digital concierges that remember your favorite popcorn
- Children's hospitals with multilingual avatars that interact with families and explain care plans
- Retail stores with account-aware avatars that eliminate tedious forms and streamline service
These aren't sci-fi fantasies—they're practical applications of AI that solve real business challenges.
Why WWT?
WWT's strength lies in its ability to architect, build, and scale complex AI solutions. By integrating multi-vendor technologies and navigating enterprise deployment needs, WWT is turning digital humans into intelligent front-ends for personalized, multimodal, and context-retaining experiences.
While WWT didn't have a booth at SIGGRAPH, its presence was felt. The session was a powerful reminder that innovation isn't just about what you show—it's about what you build, what you learn and what you share.