AI Engineer

About Canada Quant Labs

Canada Quant Labs specializes in training, quantizing, and deploying open-weight AI models tailored to Canadian Blackwell silicon, supporting sectors like legal, medical, defence, and finance that require sovereign AI without relying on external APIs. All outputs including model weights, recipes, benchmarks, and engineering logs are publicly available to foster transparency and collaboration.

Job Overview

We are seeking a hands-on AI Engineer to lead critical segments of our train-to-deploy pipeline. This position entails training and fine-tuning large-scale language models, crafting quantization recipes, validating model performance, serving models efficiently, and documenting the process thoroughly to share with the open-source community. This role is based in Victoria, BC with a hybrid work setup and offers a competitive salary plus equity.

Key Responsibilities

• Train and fine-tune open-weight models on DGX B300 clusters, including specialized domain adaptations for Canadian legal, medical, defence, and finance corporations.
• Develop and manage end-to-end quantization workflows such as W4A16 and NVFP4 using llm-compressor, perform calibration data selection, and validate quality against FP8/BF16 benchmarks.
• Create and execute evaluation harnesses to determine model shipping criteria, including reasoning tasks, instruction-following, long-context retrieval, and domain-specific benchmarks like the CanLegal Bench.
• Deploy models via vLLM with capabilities like speculative decoding (MTP), expert-parallel mixture of experts (MoE) serving, throughput and latency profiling, and kernel-level debugging to resolve performance issues.
• Publish model documentation, reproducible recipes, benchmark methodologies, engineering logs, and contribute upstream fixes to open-source projects like vLLM and llm-compressor as needed.

Required Qualifications

• Expertise in PyTorch and transformer architectures with practical experience in training or fine-tuning models comprising billions of parameters, understanding network internals beyond framework usage.
• Hands-on experience with large language model serving or quantization frameworks such as vLLM, TensorRT-LLM, SGLang, or development of quantization tools like GPTQ, AWQ, NVFP4 that have been utilized by others.
• Strong systems knowledge including CUDA profiling, memory management, tensor and expert parallelism, and performance troubleshooting to diagnose throughput regressions.
• Demonstrable engineering contributions such as public checkpoints, kernel modules, evaluation harnesses, merged pull requests, or detailed write-ups showcasing craftsmanship.
• Ability to produce clear, thorough technical documentation to serve as official product artifacts shared with the community.

Preferred Extras

• Experience in mixture of experts (MoE) training or serving, including expert parallelism and routing mechanisms.
• Familiarity with speculative decoding techniques like MTP and EAGLE, and tuning acceptance rates.
• Expertise in benchmark design with rigorous statistical analysis and contamination control.
• Background in regulated industries such as legal, medical, defence, or finance, or knowledge of Canadian data residency requirements.
• Open-source project maintainership or significant contributor status in inference or training software stacks.

Work Environment and Culture

Our small, tight-knit team operates on a hybrid schedule in Victoria, BC. We value in-person collaboration for critical reviews and launches, flexible hours otherwise, and provide dedicated compute resources for uninterrupted experimentation. Decisions and outcomes are documented formally, with individuals expected to present their results.

Compensation and Logistics

• Base salary ranging from CAD 140,000 to 190,000, with potential equity grants.
• Position located in Victoria, BC; offers relocation support.
• Work authorization primarily for Canadian citizens and permanent residents.
• Flexible start date with an immediate hiring process underway.

How to Apply

Submit a brief introduction and résumé, including links to your work artifacts such as GitHub repositories, papers, benchmarks, or model cards. The hiring process includes an introductory call, a technical deep-dive on your prior work, a paid half-day trial working on real cluster tasks, followed by an offer.