Senior Lead ML Encoder

Overview

Client is seeking a hands-on senior Machine Learning Encoder Lead to develop a shared learned customer representation for downstream GenAI and analytics products. This role will design pretraining objectives, train and evaluate encoder models, and build production-quality ML solutions using longitudinal customer data.

The ideal candidate brings deep representation-learning expertise and can independently determine whether the approach provides meaningful downstream value. Evaluation rigor is a core deliverable, including clear evidence, uncertainty assessment, and recommendations on whether to continue or stop an approach.

Core Responsibilities

• Design pretraining objectives for customer representation and embedding models.
• Train, evaluate, and refine encoder models using longitudinal transaction, sales, interaction, and behavioral-event data.
• Develop dense customer representations that downstream GenAI and analytics products can reuse.
• Build inductive representations that can model entities with limited historical data using their own features.
• Apply self-supervised, contrastive, temporal, sequence, transformer, graph neural network, or recommender-based modeling approaches as appropriate.
• Design rigorous evaluation frameworks using time-based splits, leakage detection, cold-start analysis, held-out population transfer testing, and hard baselines.
• Assess incremental downstream value of embeddings, including calibration, stability, drift, uncertainty, and subgroup performance.
• Build production-quality Python code, training pipelines, data contracts, model versioning, serving, monitoring, and reproducible workflows.
• Partner with senior stakeholders to communicate findings, risks, uncertainty, and recommendations.
• Recommend discontinuing modeling approaches when evaluation evidence does not support continued investment.

Essential Qualifications, Skills, and Technologies

• Experience personally training an encoder or embedding model and designing its pretraining objective.
• Deep expertise in representation learning, including self-supervised learning, contrastive pretraining, temporal or sequence modeling, transformers, graph neural networks, or recommender embeddings.
• Experience modeling large-scale, sparse, longitudinal event data such as transactions, claims, clickstream, customer journeys, or engagement histories.
• Experience developing inductive representations for entities with limited history.
• Strong evaluation discipline, including time-based splits, leakage detection, cold-start analysis, held-out population testing, uncertainty assessment, and hard baselines.
• Ability to assess incremental signal and downstream embedding quality through calibration, stability, drift, and subgroup analysis.
• Strong Python development experience using PyTorch or JAX.
• Experience with SQL, distributed data processing, and cloud-based model training at scale.
• Experience moving models from research into production, including pipelines, versioning, serving, monitoring, data contracts, and reproducibility.

Preferred Skills or Experience

• Experience with customer-360 representations, behavioral embeddings, recommender systems, or foundation models trained on event data.
• Familiarity with privacy, fairness, and re-identification risk for learned representations of individuals.
• Publications, patents, or public applied work in representation learning.
• Experience working with large-scale behavioral event data in consumer technology, marketplaces, streaming, financial services, payments, or advertising technology.

Work Details

• Bay Area-based hybrid role.
• Minimum three days per week onsite.