Senior AI Researcher – Relational Foundation Models

SAP IT Business Systeme•Palo Alto, CA

10h•$148,600 - $306,300

About The Position

SAP is uniquely positioned to lead the next wave of AI by infusing intelligence directly into the business processes that run the world. Our team's mission is to develop Foundation Models for structured data. We will start by launching and continuously improving SAP-RPT-1, the first version of our relational model portfolio. This is part of our commitment to lead research in a field where SAP's expertise in enterprise data provides us with a solid competitive advantage. As a Senior AI Researcher, you will improve Relational Foundation Models (RFMs). These models are general-purpose. They learn, reason, and make predictions from complex, multi-table enterprise data. They operate without the need for manual feature engineering or task-specific pipelines. Your work will span fundamental research and scalable system development, from initial hypothesis through production-ready methods that operate over real enterprise databases. We are looking for innovators ready to bridge the gap between cutting-edge AI research and the complex, structured reality of global enterprise data.

Requirements

PhD in Machine Learning, Computer Science, Statistics, Applied Mathematics, or a closely related field — or equivalent demonstrated research experience.
Strong publication record at leading venues such as NeurIPS, ICML, ICLR, KDD, AAAI, or AISTATS.
Deep expertise in modern foundation-model techniques. This includes Transformer architectures and attention mechanisms. It also includes large-scale pretraining, in-context learning, and transfer learning.
Hands-on experience in at least one of: tabular machine learning, graph neural networks, relational learning, or structured prediction.
Excellent Python and PyTorch skills with the ability to independently implement, debug, and evaluate research ideas end-to-end.
Rigorous experimental practice: strong baselines, controlled ablations, honest failure analysis, and awareness of data leakage and evaluation artifacts.
Ability to operate with substantial research independence — from identifying problems through validated results.
Architectural thinking and hands-on coding skills. You design secure, scalable systems, reason about data structures and APIs, and use observability, CI/CD and automated testing to ship safely and fast. You can dive deep on performance tuning, apply secure-by-default patterns, and mentor others in writing clean, maintainable code.

Nice To Haves

Experience with relational deep learning, graph neural networks, or heterogeneous graph learning.
In-context learning is not limited to natural-language tasks. It also covers meta-learning and amortized Bayesian inference.
Create synthetic data for model pretraining. Conduct large-scale pretraining with several datasets.
Knowledge with relational database systems, SQL, or data-warehouse architectures.
Distributed training and large-scale GPU experimentation.
Benchmark development or contributions to open-source machine learning libraries.

Responsibilities

Lead original research into Relational Foundation Models. These models are architectures that learn from many multi-table enterprise databases. They can also adjust to different schemas and prediction tasks.
Design and develop novel model architectures, pretraining strategies, learning objectives, and inference methods for relational and structured data.
Run large-scale distributed GPU experiments; build rigorous baselines, ablation studies, and evaluation benchmarks.
Contribute to datasets, synthetic data generators, and evaluation frameworks for relational learning.
Translate research findings into scalable prototypes ready for integration with engineering and product teams.
Mentor researchers and engineers and provide technical leadership on research projects.
Publish at top-tier venues (NeurIPS, ICML, ICLR, KDD, or equivalent) and contribute to the broader AI research community.
Stay current with the latest advances in foundation models, tabular learning, graph learning, and large-scale machine learning.