Senior Lead ML Encoder

Skill; San Francisco (CA); On-site; USD 125,000 - 139,000; Full time

Benefits offered by this job

subsidized health, vision, and dental

paid sick leave

retirement plans with a match

free online training through AquentGym

Job summary

Aquent Talent partners with leading biotech firms to advance data-driven medicine. This temporary, high-impact role focuses on training dense embeddings from longitudinal histories to power next‑gen generative AI and analytics across the enterprise.

You will design pretraining objectives, build encoder models, and implement robust evaluation while collaborating with stakeholders to determine when to proceed or pivot.

Qualifications

• Hands-on experience designing pretraining objectives and training encoder/embedding models from scratch.
• Deep expertise in representation learning: contrastive learning, temporal sequence modeling, Transformers, Graph Neural Networks (GNNs), or recommender embeddings.
• Proven track record modeling large, sparse, longitudinal event data.
• Experience building inductive representations that generalize to entities with minimal history.
• Strong evaluation hygiene, including time-based splitting, leakage detection, uncertainty estimation.
• Strong Python engineering with PyTorch or JAX, SQL, distributed data frameworks, and cloud-based training at scale.
• Demonstrated ability to carry models into production with data contracts, versioning, serving pipelines, and monitoring.
• Strong communication skills to articulate uncertainty and recommend stopping approaches when signal is insufficient.

Responsibilities

• Design pretraining objectives and build self-supervised or contrastive representations using longitudinal data.
• Develop, train, and evaluate end-to-end encoder models using PyTorch or JAX, with scalable distributed processing.
• Establish evaluation frameworks including time-based splits, leakage checks, and transferability assessments.
• Write clean production-grade Python to move models from research to production pipelines.
• Judge embedding signal by metrics on calibration, drift, and subgroup performance for incremental value.
• Present findings and trade-offs to senior stakeholders with data-driven recommendations.

Skills

Pretraining objectives

Encoder embeddings

Representation learning

Longitudinal data

Time-based evaluation

Python (PyTorch/JAX)

SQL & distributed data

Production ML

Communication skills

Tools

PyTorch

JAX

SQL

Distributed data

Cloud training

Job description

Overview

Placement Type: Temporary

Salary: $91.09-101.21 Hourly

Start Date: Sep 21, 2026

Aquent is proud to partner with a leading biotechnology company that is at the forefront of scientific discovery, dedicated to improving human health. This organization leverages cutting-edge research and innovative technologies to develop life-changing therapies. Join a team that is making a profound impact on patients' lives globally, driving forward the future of medicine through data-driven insights and advanced analytical capabilities.

Are you a visionary machine learning leader passionate about building foundational data representations that unlock unprecedented insights? We are seeking an exceptional individual to spearhead the design and execution of a groundbreaking, comprehensive customer representation model. In this high-impact, temporary opportunity, you will be instrumental in training a dense vector embedding on extensive longitudinal transaction and interaction history. Your work will directly power next-generation generative AI and advanced analytics products across the entire enterprise, fundamentally transforming how our client understands and serves its customers. This is your chance to pioneer a first-of-its-kind solution, driving innovation and making a tangible difference in the biotechnology space.

Key Responsibilities

• Design pretraining objectives and build self-supervised or contrastive representations using longitudinal customer transaction, sales, and interaction data.
• Develop, train, and evaluate end-to-end encoder models utilizing PyTorch or JAX, ensuring scalable distributed data processing.
• Establish robust, rigorous evaluation frameworks, including time-based splits, leakage detection, cold-start slices, and transferability to held-out populations.
• Write clean, production-grade Python code to transition models from research into reliable, versioned, and monitored training and serving pipelines.
• Judge downstream embedding signal by measuring calibration, stability, drift, and subgroup performance to determine if models deliver incremental value.
• Present research findings, model capabilities, and trade-offs to senior stakeholders, providing data-driven recommendations on whether to advance or halt specific modeling approaches.

Must-Have Qualifications

• Hands-on experience designing pretraining objectives and training encoder/embedding models from scratch (not just fine-tuning or consuming pre-trained models).
• Deep expertise in representation learning techniques, such as contrastive learning, temporal sequence modeling, Transformers, Graph Neural Networks (GNNs), or recommender embeddings.
• Proven track record modeling large, sparse, longitudinal event data (e.g., transactions, customer journeys, clickstreams, or engagement histories).
• Experience building inductive representations that generalize to entities with minimal history using their own features rather than static lookup tables.
• Exceptional evaluation hygiene, including time-based splitting, leakage detection, uncertainty estimation, and hard baseline comparisons.
• Strong engineering expertise in Python with PyTorch or JAX, SQL, distributed data frameworks, and cloud-based training at scale.
• Demonstrated ability to carry models into production, including setting up data contracts, versioning, serving pipelines, and monitoring.
• Strong communication skills with the confidence to articulate uncertainty and recommend stopping viable approaches when signal is insufficient.

Nice-to-Have Qualifications

• Direct experience building comprehensive customer representations, behavioral embeddings, or foundation models over event data.
• Familiarity with privacy, fairness, and re-identification risks in learned representations of individuals.
• Relevant publications, patents, or public applied work in representation learning.
• Prior experience working in high-volume behavioral data environments (e.g., consumer tech, financial services, payments, adtech, or marketplaces).

About Aquent Talent

Aquent Talent connects the best talent in marketing, creative, and design with the world's biggest brands.

Our eligible talent get access to amazing benefits like subsidized health, vision, and dental plans, paid sick leave, and retirement plans with a match. We also offer free online training through Aquent Gymnasium. More information on our awesome benefits!

• subsidized health, vision, and dental plans
• paid sick leave
• retirement plans with a match
• free online training through Aquent Gymnasium

Aquent is an equal-opportunity employer. We evaluate qualified applicants without regard to race, color, religion, sex, sexual orientation, gender identity, national origin, disability, veteran status, and other legally protected characteristics. We're about creating an inclusive environment-one where different backgrounds, experiences, and perspectives are valued, and everyone can contribute, grow their careers, and thrive.