3  Teaching Practice

My teaching practice is defined by a practice-first approach: I do not prepare students to do statistics “later”; I have them act as statisticians from day one. This philosophy relies on four pillars that define my pedagogical signature.

3.1 Pedagogical Signature

  1. Situational authenticity: assessments are not school exercises, but simulations of professional mandates (consultant, analyst, developer).
  2. Reproducibility by default: technical rigor (Git, Quarto) is not a peripheral skill; it is the language of scientific evidence itself.
  3. Structured collaboration: the classroom is a workshop where code is written, read, and critiqued collectively, reproducing the dynamics of a modern data team.
  4. Technological responsibility: the use of tools (AI, libraries) is encouraged but always framed by requirements for validation and transparency.

3.2 Learning Through Projects: The “Learning by Doing” Strategy

I prioritize project-based learning as the main driver of skills development. This strategy adapts across all levels, from introductory courses to specialized seminars, by adjusting the level of autonomy.

3.2.1 Immersion at Universite Laval: The Junior Data Scientist (STT-1100)

In this first-year course, the project-based approach is scripted to reduce code-related anxiety.

  • Intention: transform the perception of failure (“I am bad”) into iteration (“my code has a bug”).
  • Implementation: students take on the role of a “junior analyst” in a fictional institute. Each module is a mission (for example, “Antarctic Mission” for data cleaning) that ends with an executable deliverable.

3.2.2 Technical Projects at UQAC: Theory in Action

At Universite du Quebec a Chicoutimi, I apply the same logic in courses with more theoretical or specialized content.

  • Data Visualization and Interface Design (8INF416): instead of in-class exams, students complete a critique and redesign of an existing media visualization, justifying their choices through graphical perception theory.

  • Statistical Methods for Massive Data (8STT108): the course is built around architecture mini-projects where students must not only analyze data, but select and justify the tool that matches data volume.

  • Introduction to Data Science (8INF404): statistics course for data science (official title).

  • Pedagogical gain: this approach forces students to go beyond the “how” (syntax) and address the “why” (analysis strategy).

3.2.3 R for Scientists (STT-4230/6230)

For advanced students, the project becomes a full software development exercise.

  • Example deliverables: creation of a functional R package, interactive web app (shiny), interactive tutorial (learnr), flexdashboard, etc.
  • Professional practice: continuous integration (CI/CD) through GitHub Actions; the code must not only produce the correct answer, but pass automated tests on a remote machine.
Note

Project as alignment: Across all these courses, the project is not a peripheral activity (“if there is time”), but the core of pedagogical alignment. Lectures are resources to support project success, not the opposite.

3.3 Supporting the Next Generation of Researchers

My involvement at UQAC extends beyond the classroom; it is part of a broader commitment to building an autonomous and rigorous scientific community in the region.

Technical mentoring and reflective posture I support students in their first steps toward research or the labor market by emphasizing posture. Beyond code, I work with them on:

  • the ability to read and critique technical documentation,
  • autonomy in debugging (moving from “it does not work” to “here is the error and what I already tried”),
  • communication of results to non-specialists.

This close supervision of internships and projects helps identify talent and guide students toward graduate studies or key industry roles, strengthening the local scientific ecosystem.

3.4 Assessment and Integrity in the AI Era

The rise of LLMs (ChatGPT, Claude) has made assessments based on knowledge recall or simple code production obsolete. I redesigned my assessments to measure the process rather than only the final product.

  1. Mandatory traceability: in STT-1100, assessment includes Git commit history. A “perfect” project that appears suddenly without history is suspicious; an iteratively built project is valued.
  2. AI as a supervised partner (STT-4230):
    • I do not ask students to “avoid AI”.
    • Assessment requires integrating an LLM to generate a contextual interpretation of statistical results.
    • Success criterion: the student’s ability to preprocess the data sent to the model (minimization) and validate model output (hallucination detection).
Warning

Paradigm shift: We no longer assess only “Does the code run?”, but “Is the process robust, transparent, and verified?” Academic integrity becomes a professional verification skill.