The energy in the room at AI Demo Night #9 was unmistakable.
Ideas were shared, questioned, and stress-tested in real time. The questions were sharp, the conversations candid, and a clear theme emerged throughout the night: the future of AI will be defined by systems that hold up under real-world constraints.
Hosted by TechUnited at Stevens Institute of Technology, AI Demo Night #9 brought together founders, engineers, operators, researchers, and enterprise leaders to explore what it takes to build AI that works in production.
The Energy in the Room at AI Demo Night #9
AI Demo Night has always focused on creating space for high-signal conversations, and this one delivered.
The room included people actively building and deploying AI across healthcare, infrastructure, enterprise software, and research – with attendees from organizations such as Google, Verizon, Nokia Bell Labs, Mayo Clinic, JPMorgan, and Rutgers.
The discussion consistently centered on practical execution. Questions focused on how AI systems behave under real conditions, where they fail, and what is required to make them reliable at scale.
Live AI Demos: Stress-Testing Ideas in the Real World
Three live demos grounded the evening in real applications, each highlighting a different dimension of production-ready AI.
Predicting Sepsis with AI in Healthcare
Andy Bala, Founder & CEO of Delphine Diagnostics, presented an AI-driven system designed to predict sepsis hours before clinical symptoms appear.
With sepsis remaining one of the leading causes of death in U.S. hospitals – and up to 80% of cases considered preventable – the stakes were clear. Delphine’s system uses real-time patient data to generate early risk signals, enabling clinicians to intervene sooner, reduce ICU stays, and improve outcomes.
The conversation extended beyond the model itself to include clinical workflows, regulatory considerations, and the operational realities of deploying AI in healthcare settings.
Agentic AI for Customer Conversations and Revenue
Next, Shashank Singh, Founder & CEO of Bakstage AI, demonstrated an agentic AI platform focused on improving customer engagement and conversion.
Backstage AI shared concrete results, including an 85% reduction in no-show rates, 20+ paying customers, and partnerships with IBM and Deloitte. The system enables AI agents and humans to operate within a single interface, allowing seamless handoffs during live customer interactions.
The discussion highlighted how system design influences trust, efficiency, and real-world adoption.
Rethinking AI Infrastructure with Analog Computing
Harrison Muchnick presented a proposal for analog neural networks, exploring custom hardware circuits designed to perform AI inference at near the speed of light.
The approach demonstrated significant potential for energy efficiency, with theoretical workloads requiring a fraction of the power used by traditional GPUs. Harrison also discussed practical challenges, including accuracy drift, environmental sensitivity, and the difficulty of scaling physical components for large models.
The demo illustrated how infrastructure-level innovation requires direct engagement with physical and operational constraints.
From AI Builders to AI Architects
The keynote tied the evening’s themes together.
Nick Gu, President of Pioneering Minds AI and a senior technical leader at Google, outlined a mindset shift required to build durable AI systems. As AI-generated applications become more common, many failures stem from building without clearly defining the underlying problem, success criteria, and evaluation framework.
Nick emphasized the importance of thinking like an AI architect – starting with problem definition, evaluation-driven development, and structured iteration.
Key ideas included:
Defining success criteria before deployment
Using evaluation and labeled data to guide iteration
Treating AI as a system that requires context, feedback, and governance
Balancing public benchmarks with real user-focused evaluation
Why AI Architecture Matters More Than Ever
As AI systems move deeper into healthcare, finance, and enterprise operations, the risks and responsibilities increase.
Cost control, security, privacy, and reliability are no longer secondary considerations. Building durable AI requires systems that can be monitored, audited, and improved over time — and teams that understand how those systems behave in production.
AI Demo Night #9 reinforced the value of architectural thinking in meeting these challenges.
Why AI Demo Night Matters
Events like AI Demo Night create space for early-stage ideas to be examined openly and rigorously. They bring together builders and operators to share lessons, challenge assumptions, and collectively advance the state of applied AI.
This kind of community-driven dialogue plays a critical role in shaping innovation ecosystems and supporting responsible AI development.
Event Partners and Sponsors
We’re grateful to our sponsors – Tech Council Ventures, EY, PwC, Verizon, KPMG, Nokia Bell Labs, PNC, and NJEDA – for supporting AI Demo Night and helping make high-signal community conversations like this possible.
Looking Ahead
The future of AI will be shaped by teams who design resilient, production-ready systems and understand the environments in which those systems operate.
AI Demo Night #9 made clear that this community is focused on doing exactly that.