Be Boring. Make Money.™ is our investment approach, but we are anything but dull. We are a diverse, passionate team united by our commitment to provide our clients with financial peace of mind and our desire to create one of the best companies in the world.
A career with us will be demanding; however, the experience will also be dynamic, supportive, intellectually stimulating, and above all, meaningful. We take pride in doing things a little differently than other investment firms. Innovative ideas, diverse opinions and experiences, and teamwork are expected and rewarded regardless of title or tenure.
Dedicated to our people, we have consistently been recognized as a preferred employer, receiving awards such as Canada’s Top 100 Employers (2026), Alberta Top Employers (2026), and Canada’s Top Small & Medium Employer (2025).
Full Time, Contract (4 Months)
January – April 2027
Location: Calgary, AB
The Opportunity
Mawer is looking for an AI Engineer Student to work on the model layer of our internal AI platform training, adapting, and evaluating models that we run ourselves, on our own hardware.
We want someone who has actually stood up open-weight models locally: downloaded the weights, fought the memory budget, served them, and watched the latency numbers. Someone who has taken a course or done serious self-study on transformers and the architectures that have followed them and can explain why attention scales the way it does, what mixture-of-experts actually buys you, and when a state-space model is the better answer. And someone who wants to work on the model itself, the data, the training loop, the evaluation, rather than only calling an API someone else trained.
The work spans the full adaptation pipeline: continued pretraining to bring a base model into our domain, supervised fine-tuning to teach it a task, and reinforcement learning to shape how it behaves. That is a genuinely broad mandate for a four-month term, and we are describing it honestly.
Duties and Responsibilities
What You'll Work On
- Running models locally. Open-weight models on our own infrastructure: quantization, serving stacks, GPU and CPU memory trade-offs, throughput and latency measurement, and the practical question of what a given model actually costs to run.
- Continued pretraining. Taking a base checkpoint and adapting it to financial language and internal corpora: data curation, mixing, tokenization, and the effect of each on downstream behaviour.
- Supervised fine-tuning. Dataset construction, prompt formatting, hyperparameter selection, overfitting control, and the difference between a model that memorized the training set and one that learned the task.
- Reinforcement learning. Preference optimization and reward-based training: reward modelling, DPO-family methods, RLHF, and verifiable-reward approaches. Including how reward design goes wrong and what a reward-hacking model looks like.
- Building the measurement that tells you whether any of the above actually helped: held-out sets, regression suites, contamination checks, and the discipline to report a negative result.
- Training infrastructure. Experiment tracking, checkpointing, reproducibility, parameter-efficient methods, and enough distributed training to know when it is worth the complexity.
Key Responsibilities
- Serve open-weight models locally and characterize them: memory footprint, tokens per second, latency under load, quality against the base checkpoint.
- Build and maintain training datasets from internal and public corpora, with attention to licensing, contamination, and quality.
- Run fine-tuning experiments end to end and write up what worked, what did not, and why you think so.
- Assist with reinforcement learning experiments: reward design, preference data collection, and evaluation of trained policies.
- Design evaluations that separate real capability gains from benchmark overfitting, and apply them consistently across runs.
- Document training runs so a colleague can reproduce them from the config and the data version alone.
- Contribute to the team's emerging practice for model training and evaluation, and to the standards we are still writing.
What We’re Looking For
This is intended to be a stretch opportunity. We are describing someone we want to help grow over the term, not a checklist for a finished specialist. If you can reason well, build carefully, and learn fast, we want to hear from you.
- Hands-on experience running local models. You have downloaded open-weight weights and served them yourself - llama.cpp, vLLM, Ollama, Hugging Face Transformers, TGI, or equivalent. You know what quantization does to quality and why context length is expensive.
- Coursework or serious self-study in transformers and modern architectures. Attention, positional encoding, mixture-of-experts, state-space and hybrid models. You can explain the trade-offs rather than recite the acronyms.
- Python and PyTorch. Comfortable reading a model's source, not only calling it.
- Some experience training or fine-tuning. Even at small scale, on a single GPU or a rented one. We care that you have run the loop and watched the loss curve, not that you have trained something large.
- An instinct for data quality. You have noticed that most model problems are dataset problems.
- Clear technical writing. You can describe an experiment and its result so someone else can act on it.
- Currently enrolled in a post-secondary program and returning to full-time studies or graduating following the work term.
Valuable Experience
These are not requirements. They are the things that would make you unusually effective here.
- Reinforcement learning beyond the textbook. DPO, ORPO, GRPO, RLHF, or verifiable-reward training you have actually run.
- Parameter-efficient fine-tuning. LoRA, QLoRA, adapters, and knowing when full fine-tuning is worth the cost.
- Distributed or multi-GPU training. Even a two-GPU run you had to debug.
- Evaluation design. Building a benchmark, or discovering that an existing one was measuring the wrong thing.
- Quantization and inference optimization. GGUF, AWQ, GPTQ, speculative decoding.
- Financial or domain-specific language. An interest in how models handle numbers, filings, and precise terminology.
- A portfolio of things you built and can talk about honestly, including the ones that did not work.
Why Mawer?
- Directly contribute to the firm’s growth as a global, independent investment manager.
- Learn and work within a high-performing team that has consistently delivered exceptional long-term investment results and high levels of client satisfaction.
- Be a part of a meritocracy in which the quality of ideas supersedes hierarchy; where every role is valued, and every voice heard. At Mawer, talented team members can progress quickly, regardless of tenure or experience.
- Our unique culture, which is integral to who we are and (we think) a competitive advantage. Our culture is shaped by our core values (integrity, putting clients’ interests first, excellence, teamwork, and long-term thinking) and key behaviours (curiosity, candour, accountability, and appreciation), which are much more than words on a page. They guide the way we operate, make decisions, and communicate every day.
- We offer a flexible work environment with a blended workplace approach tailored to the unique needs of each department and team. This allows employees to seamlessly combine in-person and remote work experiences, fostering collaboration, promoting work-life integration, and empowering our team members to thrive in a dynamic, interconnected professional setting.
To apply:
We’re keen to know more about something you’ve built and trained, what it was, what you tried, what the result was, and what you would do differently. We are more interested in your reasoning than your transcript.
- Send us your resume and a short note telling us about what you’ve built and trained, what worked, what didn’t and what you might do differently if you could do it all over again.
- Submit your full application by November 25, 2026, to: https://www.mawer.com/careers/
All candidates must be eligible to work in Canada and currently enrolled in a post-secondary program and returning to full-time studies or graduating following the work term.
We believe experience comes in many forms and we’re committed to adding new perspectives to our team. If your experience is close to what we’ve outlined, we encourage you to apply.
At Mawer, we are committed to fostering a culture of diversity and inclusion. We strive to attract, hire, develop, and retain the best people from a wide variety of backgrounds/experiences and create an environment where unique ideas and perspectives are heard and valued. We believe a diverse and inclusive culture fuels innovation and enables more effective decision-making across the firm, helping to improve investment performance, strengthen client partnerships, and create more value for all stakeholders: clients, employees, owners, and society.
We administer all practices and programs without discrimination and strive to provide equal access to employment. A Human Resources representative will work with applicants who request accommodation or special assistance during the application process. If you need accommodation or assistance, please contact us at hraccommodation@mawer.com.
Our processes are people-driven. We may use technology for scheduling, communication and screening but we do not use AI for candidate assessment, or selection; those decisions are made by our team.
About Mawer
Established in 1974, Mawer Investment Management Ltd. is an independent firm that provides world-class, global investment management services to a broad range of institutional and individual clients. We are proud of our reputation as one of Canada’s most respected investment management firms and strive to earn such respect by continually delivering long-term investment excellence, developing strong client partnerships, and living our core values.
No telephone inquiries please.
We thank all applicants and advise that only those selected for an interview will be contacted.
