Senior Machine Learning Engineer

Senior Lead ML EncoderJob Title: Senior Lead ML EncoderRole Level: SeniorDuration: 12 MonthsLocation: South San Francisco, CAWork Arrangement: Hybrid — Minimum 3 days per week onsiteFocus Area: Customer Representation Learning & Encoder DevelopmentAbout the RoleWe are looking for a Senior Lead ML Encoder to build the first shared learned representation of customers.The goal is to create one dense vector representation per customer, trained on longitudinal transaction, sales, and interaction history. This shared representation will be reusable across downstream Generative AI and analytics products, eliminating the need for individual teams to independently derive their own representation of the same customer data.This is a hands-on senior-level contractor role. The ideal candidate will independently define modeling objectives, design evaluation strategies, write production-quality code, and determine whether an approach is delivering meaningful results.Evaluation is a core part of the deliverable—not an afterthought.Key ResponsibilitiesDesign and implement pretraining objectives for customer representation learning.Develop, train, and evaluate customer encoder and embedding models.Work with large-scale, sparse, longitudinal event data including transactions, sales, and customer interactions.Develop inductive representations that can represent customers with limited history using available features rather than relying solely on lookup tables.Design rigorous evaluation frameworks using time-based splits, leakage detection, cold-start analysis, held-out populations, and strong baseline comparisons.Assess whether learned embeddings provide genuine incremental downstream signal.Evaluate model calibration, stability, drift, uncertainty, and subgroup performance.Build scalable ML pipelines and production-quality modeling code.Partner with data and engineering teams to establish data contracts, training pipelines, model versioning, serving, monitoring, and reproducibility.Present modeling results, limitations, and uncertainty clearly to senior stakeholders.Make data-driven recommendations about whether to continue, modify, or stop an approach that is not producing sufficient value.Minimum QualificationsDemonstrated experience personally training encoder or embedding models, including designing the pretraining objective—not simply consuming pretrained embeddings or fine-tuning existing LLMs.Deep expertise in representation learning, including one or more of:Self-supervised learningContrastive learningSequence modelingTemporal modelingTransformersGraph Neural Networks (GNNs)Recommender-system embeddingsExperience modeling large-scale, sparse, longitudinal event data such as transactions, claims, clickstream data, customer journeys, or engagement histories.Experience building inductive representations for entities with limited historical data.Strong understanding of rigorous ML evaluation, including time-based data splits, data leakage detection, cold-start evaluation, held-out population testing, hard baselines, and uncertainty estimation.Ability to determine whether embeddings provide meaningful incremental downstream signal.Experience evaluating calibration, stability, model drift, and subgroup performance.Strong Python engineering skills.Hands-on experience with PyTorch or JAX.Strong SQL skills and experience with distributed data processing.Experience training models at scale using cloud-based infrastructure.Experience taking ML models from research into production, including training pipelines, versioning, serving, monitoring, data contracts, and reproducibility.Strong communication skills with the ability to present technical findings and uncertainty to senior stakeholders.Ability to independently challenge assumptions and recommend stopping an approach when evidence indicates it is not working.Preferred QualificationsExperience with Customer 360, customer representations, behavioral embeddings, or recommender systems.Experience developing foundation models or learned representations over event data.Familiarity with privacy, fairness, and re-identification risks associated with learned representations of individuals.Publications, patents, or publicly available applied work in representation learning.Experience working with large-scale behavioral data in industries such as:Consumer technologyMarketplacesStreamingFinancial servicesPaymentsAdvertising technology