ML Encoder Lead

About the role

Proclinical is seeking an ML Encoder Lead to focus on customer representation learning and encoder development. The role involves building a shared learned representation of customers using longitudinal transaction, sales, and interaction data. This representation will support downstream generative AI and analytics products. This is a hands-on role requiring expertise in model design, evaluation, and production-level coding.

Responsibilities

• Design pretraining objectives for customer representation learning.
• Train and evaluate encoders, ensuring rigorous testing and validation.
• Develop models using representation learning techniques such as self-supervised or contrastive pretraining, transformers, and graph neural networks.
• Handle large-scale, sparse, longitudinal event data (e.g., transactions, clickstreams, customer journeys).
• Build inductive representations for entities with limited historical data.
• Conduct thorough evaluations, including time-based splits, leakage detection, cold-start scenarios, and transfer to held-out populations.
• Assess embedding quality for downstream applications, focusing on calibration, stability, drift, and subgroup performance.
• Write production-level Python code using frameworks like PyTorch or JAX, and manage distributed data processing and cloud-based model training.
• Transition models from research to production, including data contracts, training pipelines, versioning, serving, monitoring, and reproducibility.
• Present findings and uncertainties to senior stakeholders and recommend adjustments or cessation of approaches when necessary.

Requirements

• Proven experience in training encoders or embedding models, including designing pretraining objectives.
• Expertise in representation learning, sequence modeling, and temporal modeling.
• Strong background in handling large-scale behavioral event data.
• Proficiency in Python, PyTorch or JAX, SQL, and distributed data processing.
• Experience with cloud-based model training at scale.
• Familiarity with rigorous evaluation practices and ability to assess embedding quality.
• Ability to communicate technical findings effectively to diverse stakeholders.

Preferred qualifications

• Experience with customer-360 representations, recommender systems, or foundation models over event data.
• Knowledge of privacy, fairness, and re-identification risks in learned representations.
• Publications, patents, or public work in representation learning.

Pay

$80 to $120 per hour.