Plenary Talks

Abstract: This talk will cover the evolution of the tech and the uses and investments in Canada as well as the shift in the AI Strategy the Task Force has been focused on.

Abstract: AI is transforming the game lifecycle by improving design workflows, enabling more personalized and safer player experiences, and helping engineering teams build, test, and deploy games more efficiently.

Abstract: Navigating a busy hospital requires more than just obstacle avoidance—it requires social intelligence. In this talk, I will discuss the complexities of high-traffic hospital corridors, and broader future directions for social navigation in unstructured human environments.

Abstract: The talk will cover the following projects

- our collaboration with Fields of the World, a PyTorch model and training/verification data set for producing field boundaries from satellite data, at scale (300M fields, globally).

- quick introduction of Google DeepMinds' AlphaEarth Foundation Model. 

- how we are using FTW and AEF to predict soil biology, following Miraterra's acquisition of Trace Genomics, in collaboration with Wherobots.

- how this approach to soil microbiology analytics is integrated into our digital agronomy LLM. 

Abstract: The BC Cancer Registry (BCCR), part of the Provincial Health Services Authority, processes over 5 million pathology and imaging reports annually to identify and register new cancer cases across British Columbia. Beginning in 2021, BCCR partnered with the University of British Columbia's Data Science Institute to develop and deploy language models that automate key registry workflows, from identifying reportable cancers in unstructured pathology text to extracting structured tumour data. This collaboration produced peer-reviewed research, operational AI tools now running in production, and a domain-adapted transformer model successfully adapted for use by another Canadian province with over 99% recall for cancer detection. These tools have improved efficiency, reduced time to data availability, and freed subject matter experts to focus on complex cases. This talk will describe how the partnership worked in practice and lessons learned in translating academic AI research into deployed health system tools.

Abstract: Large language models and agentic AI systems are rapidly reshaping how organizations think about automation, decision support, and knowledge workflows, yet many enterprises struggle to move from experimentation to governed, production-grade solutions that address real operational challenges. To bridge this research-to-adoption gap, ProCogia designed the Founders Cup AI competition as a structured innovation program that combines practitioner domain expertise, graduate student enablement, and external industry validation to accelerate the development of practical generative AI solutions.

The initiative brought together consultants from data engineering, analytics, AI, and business consulting to articulate concrete problem statements drawn from client projects and internal operations, ranging from HR process bottlenecks to contact center overload and secure enterprise AI adoption barriers. Multidisciplinary teams were paired with graduate students and academic participants, who supported rapid experimentation with modern GenAI tools, prototyping workflows, and evaluation techniques, while ProCogia consultants ensured that emerging concepts remained grounded in regulatory, operational, and change-management realities. Shortlisted prototypes were evaluated in a final round by external industry leaders, who scored each solution on technical robustness, domain fit, adoption feasibility, and expected business impact rather than novelty alone. Through this process, several initiatives progressed beyond the competition into active solution development: 

• Haigent, a suite of agentic HR operations capabilities that connect to enterprise systems to 

automate onboarding, benefits, and employee support; 

• Dataegis, a private, secure, and governed multi-model AI environment that allows enterprises to safely use LLMs against internal data; and 

• CallYeah, AI voice agents for industry-specific customer service workflows that integrate with real systems to resolve requests in real time.

Taken together, the Founders Cup AI competition demonstrates a repeatable model for industry–academia collaboration in applied GenAI: real-world problem curation by practitioners, enablement.

Abstract: LLMs are increasingly being used in legal settings to streamline research and improve access to the law for non-experts. Yet hallucinations remain a serious concern in this high-stakes domain, where errors can carry significant consequences. In this talk, I will introduce a new benchmark of legal queries and answers grounded in Canadian case law, and share findings on how well LLM-based tools perform compared to law students and paralegals. I will then focus on a key driver of hallucinations: model overconfidence, particularly when responding to underspecified queries. I present our InfoGatherer framework, which integrates LLMs with probabilistic graphical models to reduce uncertainty through targeted follow-up question generation.

Abstract: I argue that functionality should serve as the foundational layer of intelligence for AI systems operating in the physical world, centering on what things do, how they work, and the ultimate goal of generating, and eventually building, physical entities that function as intended. I will briefly cover our research along this pursuit, then ground it in my work at Augmenta, a Canadian startup, for the architectural, engineering, and construction (AEC) industry.

Abstract: I will briefly overview some projects I have undertaken using machine learning with BC companies over the last ten years, and then talk about some of our research on understanding and improving the algorithms used to train machine learning models.