MSc or PhD in a quantitative field (statistics, CS/ML, applied math, physics, or similar).
Deep grounding in probabilistic modeling and statistics, with a focus on estimators and uncertainty.
Demonstrated research ability through publications, strong thesis, or developed algorithms.
Experience with limited or no labeled data using weak supervision or unsupervised methods.
Graph modeling experience, particularly in graph ML or network analysis.
Strong Python skills with a focus on reproducible prototypes.
Intellectual honesty in evaluating ideas based on evidence.
Responsibilities
Take ownership of research problems from literature review to production handoff.
Design statistical models for sparse, high-cardinality event data.
Develop scoring methods for graph-structured data, focusing on hierarchies and relationships.
Build synthetic data generation strategies when labeled examples are absent.
Harden models against manipulation by anticipating adversarial influences.
Ensure model outputs are explainable and verifiable by domain experts.
Conduct rigorous evaluations to determine the viability of approaches and contribute to invention disclosures.
Benefits
Opportunity to work on cutting-edge research and development.
Collaborative environment with a focus on innovation.
Potential for patent contributions based on novel work.
Access to advanced tools and technologies in machine learning.
Engagement with a team of experts in a dynamic field.
Full Job Description
What you'll do
Take ownership of research problems end-to-end: literature review, hypothesis, prototype, empirical validation, shadow-mode evaluation, and handoff to production.
Design statistical models for sparse, high-cardinality event data - probability estimation, smoothing and backoff, calibration, and drift handling.
Develop scoring methods over graph-structured data: hierarchies, peer structures, and entity relationships.
Build synthetic data generation and negative-sampling strategies where labeled examples don't exist.
Harden models against manipulation: reason about how an adversary could influence training data, and design defenses.
Make model outputs explainable by construction - every score must decompose into a statement a domain expert can verify.
Evaluate rigorously: design metrics, calibration audits, and shadow-mode comparisons that decide whether an approach ships or dies.
Contribute to invention disclosures when your work produces novel, patentable mechanisms.
Requirements Required Qualifications
MSc or PhD in a quantitative field (statistics, CS/ML, applied math, physics, or similar).
Deep grounding in probabilistic modeling and statistics - you can reason about estimators, uncertainty, and calibration, not just fit models.
Demonstrated research ability: publications, a strong thesis, or shipped algorithms you designed from scratch.
Experience working with limited or no labeled data: weak supervision, synthetic data, self-supervision, or unsupervised methods.
Graph modeling experience: graph ML, network analysis, or probabilistic models over relational structure - our data is inherently graph-shaped.
Strong Python and solid engineering habits - your prototypes are reproducible and one step from production, not throwaway notebooks.
Intellectual honesty: you kill your own ideas when the evidence says so, and you can tell noise from signal.
Strong Advantages
Background in adversarial ML, robustness, or an adversarial application domain (security, fraud, abuse).
Work with language-model-based pipelines under statistical quality controls.
Familiarity with information-theoretic methods (e.g., surprisal-based scoring) or language-modeling techniques applied outside NLP.
Experience with explainability methods or calibrated scoring in high-stakes decision systems.