AI Governance by Design
I help organizations adopt AI they can explain, audit, and defend.
I am Daniel S. Demoz, an AI governance practitioner in Toronto. I write AI policy, assess AI risk, and review AI vendors against frameworks like NIST AI RMF, OSFI E-23, and PIPEDA. I also build machine learning systems, so my governance work starts from how models really behave.
What I do
My work is AI governance. The goal is simple: people can explain what an AI system does, show that it works, and stay accountable for the result.
AI policy, risk, and oversight
I help organizations set AI policy, assess risk, review AI vendors, and design human oversight that works in practice. I wrote Evidence-Based AI, published two SSRN papers on oversight and governance maturity, and am working toward the IAPP AI Governance Professional (AIGP) certification.
What it gives organizations: AI they can explain, audit, and defend.
Machine learning and applied AI
I build machine learning systems end to end: entity resolution, language model pipelines, forecasting, and validation. At the University of Ottawa, I led a team of four that resolved 23 million public funding records into about 851,000 organizations, and I wrote the matching, language model, and validation stages.
What it gives my governance work: I know where risk enters a real pipeline, because I build them.
Predictive analytics, big data, and visualization
I have taught neural networks, predictive modelling, big data analytics, and data mining at the postgraduate level. I have also delivered advanced statistical and predictive work for clients, including a lead-scoring model for a real estate firm.
What it gives my governance work: I can test a model's evidence, and I know when a strong number hides a weak result.
Social and management research
My training began in social and management research, with a master's in development management (First Class Honours) and field research in international development.
What it gives my governance work: I ask who is affected by an AI decision before I ask how accurate the model is.
Selected work
Each project is labelled by its actual status and scope.
Featured project · April 2026
University of Ottawa · AI for Accountability Hackathon · Team lead
Governance built into the pipeline
I led a team of four that built a platform to link organizations across Canadian charity filings, federal grants, and provincial open data. I wrote the matching, language model, and validation stages.
Governance controls are part of each stage, not added after the results.
Governance lesson: traceability, clear limits, and review steps need to be designed into the pipeline before any result is shown.
Independent prototype
Measures governance maturity across nine dimensions, including accountability, privacy, fairness, and security. It returns the main gaps, their priority, and a roadmap.
Governance focus: clear criteria and a practical improvement plan.
Independent prototype
Helps smaller organizations compare AI tools against their needs, budget, data, and governance requirements.
Governance focus: documented trade-offs, not a single unexplained answer.
Independent prototype
Ranks possible AI use cases by value, feasibility, and risk. Value is shown as a range, and risk is scored separately, so a high return cannot hide a high risk.
Governance focus: keep expected value apart from uncertainty and risk.
Applied machine learning
Predicts Toronto transit delays from public data, with route options and a station map.
Data visualization
An interactive dashboard for exploring Toronto transit performance and ridership data.
More technical work
More projects on GitHub, including NLP, churn, segmentation, SQL, and predictive modelling.
The three AI assessment tools are self-built portfolio prototypes. They are not certified products and do not replace legal advice, independent validation, or a formal compliance assessment.
My method
I developed this framework to answer six questions before any AI tool is chosen. My book, Evidence-Based AI, sets out the evidence discipline behind it. Each question produces a written record, so every decision can be checked later.
What outcome needs to improve, and how will we know it has?
OutputProblem statement and success measureWhat data exists, where does it live, and is it reliable enough to base decisions on?
OutputData inventory and quality checkWhich risks and rules apply before anyone relies on the system?
OutputRisk assessment and the rules that apply, such as PIPEDA, Quebec Law 25, or OSFI E-23 for federally regulated financial institutionsWhat is already in use, what works, and which vendor fits?
OutputVendor review and decision recordWho uses the system, who reviews its outputs, and what training do they need?
OutputOversight roles and training planWhat does success look like in numbers?
OutputBaseline measures, then reviews at 30 and 90 daysThe framework aligns with the Govern, Map, Measure, and Manage functions of the NIST AI Risk Management Framework.
AI use-case inventory · risk assessment · control matrix · vendor due-diligence checklist · model or system card · human oversight procedure · approval record · monitoring plan.
Claims need records and tests behind them. A framework mapping or an assessment score alone does not prove that an AI system is safe, fair, lawful, or effective.
Research and writing
My research asks how AI governance and human oversight work in real institutions.
Independent Researcher · ORCID 0009-0006-3072-7309 ↗
Book
A practical method for AI decisions that hold up, covering governance and alignment with Canadian privacy and AI regulation.
View on Amazon.ca ↗SSRN working paper · October 2026
Reviews AI governance across Ontario's 24 publicly funded colleges.
Read on SSRN ↗SSRN paper · August 2026
Looks at where human oversight requirements fall short in practice, and how to reduce automation bias.
Read on SSRN ↗Selected articles
Why human oversight needs clear limits, real authority, and the power to stop a system.
Read article ↗A model can be 92 percent accurate and still fail to improve the business.
Read article ↗Applying the discipline of evidence to AI decisions.
Read article ↗Experience and education
BRUKD Consultancy, Toronto
Seneca Polytechnic · Centennial College · George Brown Polytechnic · Computek College, Toronto
International development research and monitoring, including UNDP-funded research on early warning systems for coastal communities.
Contact
I am open to roles in AI governance, responsible AI, and AI risk. Email is the best way to reach me.
Toronto, Ontario