Senior Machine Learning Data Engineer
Location: Canada
The Opportunity
This is an opportunity to work alongside a highly accomplished AI research team tackling fundamental problems in advanced machine learning.
Rather than maintaining established pipelines, you’ll be helping develop new approaches to training-data engineering and curation where established playbooks often do not yet exist.
The organization is building a substantial technical team in Montreal, and we are hiring multiple people across this area.
If you’ve worked on LLM training data, large-scale NLP pipelines, foundation-model infrastructure or web-scale data processing, I’d be interested in speaking with you.
We’re working with an ambitious AI research organization building next-generation machine learning systems and are looking for multiple Senior Machine Learning Data Engineers to join its growing technical team.
This role sits at the intersection of data engineering, machine learning, NLP, and large-scale training data curation.
You’ll be responsible for building and scaling the infrastructure that transforms raw, web-scale data into high-quality datasets used to train advanced AI models.
The core challenge is engineering data quality at enormous scale: developing filtering systems, model-based quality scoring, dataset transformations, contamination detection, and tooling that allows researchers to understand and work with training corpora effectively.
What You’ll Work On
- Design, build and scale pipelines that transform raw web-scale data into high-quality datasets for training large machine learning models.
- Process extremely large unstructured text corpora, including datasets at trillion-token scale.
- Build sophisticated data-processing systems covering: Deduplication, Model-based quality scoring, Heuristic filtering, Toxicity and content-safety filtering, PII detection and removal, Metadata extraction, Dataset transformations, Versioning and provenance tracking
- Develop automated data-quality systems using a combination of heuristics, ML classifiers, LLM-based evaluators and human-in-the-loop workflows.
- Build monitoring and guardrails to identify data-quality regressions before they impact downstream model training.
- Work closely with AI researchers to understand evolving training-data requirements and identify gaps within existing datasets.
- Design robust evaluation contamination and data-leakage detection systems.
- Build internal tools that allow researchers to easily explore, query and understand large datasets.
- Optimize large-scale distributed processing systems for throughput, reliability and infrastructure cost.
What We’re Looking For
- 5+ years of experience across machine learning engineering, data processing, NLP, data infrastructure or related areas.
- Experience working with very large unstructured text datasets, ideally at web or foundation-model scale.
- Strong production-level Python.
- Hands-on experience with distributed data-processing technologies such as: Spark, Ray, Flink
- Experience designing and optimizing high-throughput distributed pipelines.
- Experience with areas such as: PII scrubbing, Content-safety filtering, Dataset quality, Evaluation contamination, Training-data governance
- Experience with workflow orchestration frameworks such as Airflow, Prefect or Dagster.
- Ability to work closely with both research and engineering teams and translate research requirements into scalable infrastructure.
Particularly Relevant Experience
We’d be especially interested in people who have worked on:
- Pretraining or post-training data pipelines for LLMs
- Foundation-model training infrastructure
- Web-scale text processing
- Common Crawl or comparable large-scale datasets
- Dataset curation for large language models
- LLM-based data-quality evaluation
- ML classifiers for filtering or scoring training data
- Large-scale data acquisition
- Dataset deduplication and contamination detection
- Distributed ML/data infrastructure
Experience with vLLM, SGLang, Docker, Kubernetes, infrastructure-as-code, experiment tracking or open-source NLP/data tooling would also be valuable.
