Innovation Network Canada
Engineering trust into the technologies that shape the future.
Who we are
Innovation Network Canada is a Canadian nonprofit that takes on hard problems at the frontier of emerging technologies. The work is to design solutions that are technically rigorous, governable by the institutions and communities that use them, and engineered for the conditions of the real world rather than for ideal infrastructure.
We partner closely with the institutions, communities, and research organizations we work with. We co-develop with the people who will use what we build. We ground every initiative in technical architecture rather than in policy aspiration. We embed sovereignty and accountability into the architecture from the outset, rather than treating them as properties to be retrofitted later. The result is technology that performs in the settings where it is needed and remains accountable to the people it serves.
We co-chair the Dynamic Coalition on Emerging Technologies at the UN Internet Governance Forum, with partners including the World Privacy Forum, OECD, UNESCO, and the UN Joint Staff Pension Fund — alongside collaborators from CERN.
Our Canadian sovereign AI
The capability that underpins our current initiatives is a sovereign AI architecture that is fully Canadian intellectual property. It is designed to operate where data sovereignty, sensitivity, or infrastructure constraints make conventional cloud-based AI inadequate or unacceptable. Sovereignty in AI, in our framing, is a property of where data resides, where computation happens, and what crosses a boundary between institutions. Our architecture addresses each of these in a complementary pair of layers — on-device computation and federated learning across institutions — that together cover the full sovereignty problem.
The architecture is sector-agnostic. It is currently most developed in clinical diagnosis and disease risk prediction, which is the proving ground for the healthcare initiative described below. The same architecture applies wherever distributed, privacy-preserving intelligence is required — environmental monitoring, genomic research, financial services, supply chain, or any other domain in which the data has constraints on where and how it can be processed.
Layer one · On-device computation
The first layer addresses the case where data should not leave the institution that holds it. A set of compact, purpose-trained models runs on the device where the data resides — a laptop in a clinic, a workstation in a sequencing facility, a tablet in a remote setting. The models are developed in Canada, without dependency on OpenAI, Anthropic, Google, or any other foreign foundation model provider. They operate without internet connectivity. No record crosses a network boundary in the course of inference.
The architecture supporting this layer is a multi-agent system in which most components are constrained to retrieval, validation, and routing rather than generation, with a single component permitted to produce reasoned output. This separation permits each output to be traced to a specific knowledge source, and allows the system to run on modest hardware without sacrificing rigour. Reference material required by the on-device model is held in a local vector database on the device itself.
Layer two · Federated learning across institutions
The second layer addresses the case where institutions need to learn from one another — where a model improves by drawing on data distributed across multiple research centres, hospitals, or communities, without any of those parties surrendering data to a central holder. For this layer, we are collaborating with Dr. Luigi Serio, the inventor of CAFEIN and head of CERN’s Technology Department.
The principle of CAFEIN is precise. When institutions participate in a federated model, the only material that crosses the network is what their local models have learned — gradient updates, parameter adjustments, the components from which the shared model is reconstructed at each round of training. The underlying data — the genome, the patient record, the imaging file — remains at the source. The shared model improves. The sources are never aggregated.
CAFEIN is one of the few implementations of this principle in production rather than research demonstration. The platform is currently deployed across a federation that includes the International Agency for Research on Cancer, Gemelli Hospital in Rome, Vall d’Hebron in Barcelona, KU Leuven, and additional partner institutions in Italy, Switzerland, Belgium, Spain, and Greece. Two EU Horizon-funded projects use CAFEIN as their underlying infrastructure: TRUSTroke, which coordinates stroke risk prediction across more than eleven thousand patients, and UMBRELLA, which addresses post-stroke care optimization. A parallel breast cancer screening initiative through IARC operates on a CAFEIN federation built on a five-hundred-thousand-individual dataset. Beyond healthcare, CAFEIN is deployed for the World Health Organization on a supply chain intelligence platform, for the World Food Programme on fraud detection in cash-based humanitarian assistance, and for the Swiss Tropical and Public Health Institute on malaria control modelling. The platform was originally developed at CERN for monitoring the LHC accelerator complex; the healthcare and humanitarian applications followed from the same underlying infrastructure.
CAFEIN’s edge variant, Piccolo Edge, is designed to run on Raspberry Pi-class hardware, with the cost of national participation on the order of one thousand Swiss francs per node. Sovereignty at this architecture is not a function of data centre capacity. It is a function of architectural choice.
Properties of the combined architecture
Our active initiatives
Indigenous Sovereign Healthcare — Tsleil-Waututh Nation Wellness Centre
In partnership with the Tsleil-Waututh Nation, we are deploying our sovereign AI architecture as the foundation for a community-controlled health system that is intended to serve as a global reference model. Our digital health platform, PHIP (Personalized Health Information Platform), delivers clinical intelligence — diagnosis, disease trajectory prediction, and prevention optimization across diabetes, cardiovascular disease, chronic kidney disease, and dementia — on-device, with no health data leaving Nation-controlled infrastructure. The work is governed by a framework grounded in the First Nations Principles of OCAP and in Canada’s obligations under the United Nations Declaration on the Rights of Indigenous Peoples. We are also involved in the development of the first By-Indigenous, For-Indigenous sovereign hospital in Canada — a health facility designed from the ground up according to Indigenous governance principles.
Governance Testbed for Trusted AI Systems
We are developing what is, to the best of our knowledge, the first AI governance testbed of its kind. Modern AI systems autonomously query databases, invoke tools, and coordinate with other agents at machine speed — yet existing governance frameworks were not designed for this. Our testbed embeds governance controls at the protocol and execution layer of AI systems, rather than appending them through organizational policy. We have discussed this work directly with the Privacy Commissioner of Canada.
Global Deep Skilling Program in Quantum and Exponential Technologies
We are building a globally accessible deep-skilling program in quantum computing, AI, and adjacent emerging technologies — in partnership with UNESCO’s International Year of Quantum Science and Technology and IEEE Quantum. The program is designed both to recruit quantum scientists internationally and to upskill domain experts in chemistry, biology, engineering, finance, and mining into quantum-literate leaders in their fields.