Skip to content

How We Work

What it’s like to be part of this lab: values, expectations, and what you’ll actually be working on.

What we work on

We build deep learning methods, often motivated by the challenge of monitoring and preserving Earth’s biodiversity. With biologists, ecologists, and taxonomists, we develop large-scale multimodal datasets and foundation models that integrate images, DNA barcodes, and text. Recurring research threads include generative modelling, graph neural networks, multimodal learning, uncertainty quantification, and the design of LLM-based agents as collaborators in research workflows.

A unique part of this lab: you’re working alongside scientists who do real fieldwork. If you’re excited about getting outside the computer-science silo (and maybe getting outside, period), you’ll be at home here.

Who we’re looking for

If you bring strong fundamentals in deep learning and any of the following experience, you’ll find a lot of room to contribute:

  • Computer vision
  • Genomics or sequence modelling
  • Foundation models and multimodal learning
  • LLMs, agents, or AI for science workflows
  • Genuine enthusiasm for working with people whose expertise lies outside ML, especially biologists and ecologists, but also philosophers, economists, and others

We run projects across the spectrum from purely methodological work to deeply applied collaborations. There’s room for both.

What we ship

We’re a remote-friendly ML research lab with a strong builder culture. We publish, we release artifacts, and we try to make work that survives contact with reality.

A sense of what we ship:

Group size

We are selective about size. Our active group sits around 10 people: large enough to have a real research community, small enough that you work directly with me and not a layer of senior students.

What we value

  • Curiosity and technical depth: wanting to understand how systems actually work.
  • Impact: outcomes in science, tooling, and real users, not just paper count.
  • Good taste: choosing problems, methods, and evaluations that matter.
  • Precision: in writing, code, experiments, and claims.
  • Ownership: you take responsibility for outcomes and communicate early when blocked.
  • Kind, no-blame culture: we fix problems; we don’t hunt for villains.

AI-first (what it means here)

We use LLMs heavily for coding, drafting, reviewing, brainstorming, and debugging. But we treat outputs as hypotheses.

  • You verify behaviour with tests, small experiments, and careful reading.
  • You document what the model helped with when it matters.
  • You develop the skill of spotting “confident nonsense” quickly.

Remote-friendly

You work where it works best for you. We don’t clock people in and out, and we don’t measure productivity by hours visible on campus.

The lab has a sunny physical space at Guelph, with windows and plants, for people who like working alongside others in person. Many team members are based in Guelph; others are scattered across southern Ontario and beyond.

A few practical things this means:

  • We default to written communication. Slack is our primary channel, and we use it heavily.
  • We hold a weekly hybrid lab meeting that brings the whole team together in person and over Zoom.
  • Real-time overlap with the rest of the team matters, so being in a distant time zone is not a good fit.
  • Being based outside Guelph is fine if you’re a strong fit and the time-zone math works.

Tooling expectations (baseline)

We’re not dogmatic about specific tools, but we have a strong preference for workflows that scale.

Expected comfort (or rapid ramp):

  • Git and PR-based collaboration
  • Linux basics
  • Reproducible environments (containers and/or modern Python packaging)
  • Writing code others can run
  • Basic CI/testing habits

Often used in this lab (examples):

  • Deep learning: PyTorch (and sometimes TensorFlow/JAX)
  • Experimentation: clusters + schedulers
  • Model serving/local inference: vLLM, Ollama, and similar tools
  • APIs: RESTful services; OpenAI-compatible endpoints where appropriate
  • Writing: clear Markdown + reproducible docs

If you actively dislike these workflows and don’t want to learn them, this won’t be a good fit.

How projects run

  • We work in small increments. A good week has multiple concrete outputs.
  • We favour prototypes early. Fast, ugly, informative.
  • We write down assumptions. Then we try to break them.
  • We keep experiments honest. Avoid leakage, overfitting to benchmarks, and “narrative-first” results.

A piece of work is “done” when:

  • Another person can run it.
  • The results are labelled, logged, and interpretable.
  • The claims match the evidence.
  • The repo/report has a minimal README explaining what it is and how to reproduce.

Feedback culture

  • Feedback is direct, kind, and specific.
  • We critique work, not people.
  • We expect revisions.
  • If you’re stuck, you say so early. Being blocked is normal; staying silent isn’t.

When things go sideways

Most experiments don’t work the first time. Most ideas need three or four tries. Negative results are the default, not a setback, and we’re a lab that takes that seriously.

What we expect when something doesn’t work:

  • You debug it. You read the code, the logs, and the literature. You form a theory about why it failed and what you’d try next.
  • You bring that theory to the next conversation. “The experiment didn’t work” is a starting point, not a deliverable. “The experiment didn’t work, here’s what I think happened, here’s what I’d try next, what do you think?” is where useful conversations begin.
  • Calibrate when to escalate. Not at the first failure, but well before you’ve burned a month going in circles.

The same instinct applies when systems break (data loss, pipeline failure, server outage). We ask what happened, what the fix is, and how to prevent recurrence. We avoid blame-based storytelling and build robust systems.

Entry points

We recruit at multiple stages. Here’s what each one looks like:

  • Undergrad to Master’s: You finished an undergraduate degree and you’re joining the lab to work on a thesis-based Master’s. We look for strong fundamentals, willingness to learn, visible building habits, and careful writing.

  • Master’s to PhD: You finished a Master’s (here or elsewhere) and you’re starting a PhD with us. We look for clear research taste, the ability to design convincing evaluations, and independent project momentum.

  • PhD to postdoc: You finished a PhD and you’re joining as a postdoc. We look for someone ready to drive an independent research thread, mentor more junior trainees, and start defining their own direction.

Where alumni go

Our alumni have gone on to roles at Anthropic, Google DeepMind, Meta FAIR, Apple, NVIDIA, Modular, Cerebras, the Vector Institute, and faculty positions in Canada. See Alumni for the full list.

The broader ecosystem

Joining this group also connects you to:

  • The Vector Institute community in Toronto: talks, compute, and a national AI research network.
  • CARE-AI at Guelph: events and interdisciplinary work on responsible and ethical AI.
  • Next AI seminars and events through my involvement as Academic Director. (Next AI itself is a separate, application-based program; some lab members have served as Scientists-in-Residence.)

What you can expect from me (the PI)

  • Clear priorities and fast feedback when it matters.
  • High standards for precision and reproducibility.
  • A bias toward autonomy: I’ll help you get unblocked, but you’ll own your work.
  • A culture that values work-life balance; intense pushes happen, but they’re the exception.

If this sounds like your kind of weird

Then you’ll probably love it here.