Senior ML Data Processing Developer
Montréal-based nonprofit AI research organization developing safe-by-design, non-agentic AI through its Scientist AI approach.
Funding history
About LawZero
LawZero conducts AI safety research and develops technical solutions for highly capable safe-by-design AI. Its Scientist AI research seeks transparent, evidence-based prediction without goal-directed behavior, with potential applications in scientific discovery and oversight of agentic AI systems.
Skills
About the Role
You will own the end-to-end pipeline that transforms raw web-scale data into high-signal datasets used to train AI models. You will design algorithmic filtering, build model-based scoring, guard against evaluation contamination, and implement novel data transformations. You will work closely with Research, Legal and Governance teams, build monitoring and internal tooling, and help acquire large-scale text corpora.
Requirements
- Degree in computer science, software engineering, or a related field
- Proven track record handling trillion-token scale unstructured text datasets, with 5+ years in data processing, ML engineering or NLP
- Hands-on experience with distributed processing frameworks (Spark, Ray, Flink)
- Experience with PII scrubbing, content-safety filtering, and evaluation-contamination prevention
- Ability to work across Research, Engineering, and Legal/Governance teams
- Strong Python proficiency with production-grade data-processing code
- Experience with pipeline orchestration frameworks (Airflow, Prefect, Dagster)
Responsibilities
- Partner with the Research team to define, build, automate, scale, and manage data pipelines for training datasets
- Build and maintain data processing pipelines including deduplication, model-based quality scoring, heuristic filtering, toxicity removal, PII scrubbing, metadata extraction, and proprietary transformations with dataset versioning and provenance tracking
- Ensure ingested data meets compliance requirements, Data Governance policies, and legal obligations
- Develop and refine the scoring and filtering toolchain including heuristics, LLM-as-a-judge evaluators, ML classifiers, and human-in-the-loop review workflows
- Instrument pipelines with data-quality monitoring, guardrails, and alerting
- Identify and acquire large-scale text corpora, conduct coverage analyses, and work with Legal & Governance to license new data sources
- Design and maintain leakage detection mechanisms to prevent evaluation contamination
- Build internal tooling and interfaces for researchers to explore and query datasets
