Principal Data & Applied Scientist – Ontologies, Knowledge Graphs & AI Research

Description

Teach AI systems how business works, so agents can act on company data and not just read it.

Agents that release blocked invoices, re-plan supply or prepare the financial close must know what an order, supplier or cost center is, how these connect, which rules and permissions apply, and how similar past decisions turned out.

For mission-critical processes, being right 90% of the time falls short.

The company's knowledge graph for its ERP platform covers 452,000 tables, 80,000 CDS views (semantic data models) and 7.3 million fields. Much of the meaning, especially in each customer's extensions, configuration and decision history, is still implicit.

The same company can appear as a customer in sales, a business partner in finance and an account in CRM, and an agent weighing a credit limit has to recognise that. This is a knowledge-representation problem more than a data-cleaning one.

The semantic layer agents depend on is mostly hand-built, one company at a time. The aim is to learn it.

The company has served organizations for over 50 years, touches more than 20 industries and is part of processes behind 80% of global commerce.

The principal data scientist sets and leads the research toward this goal and works with product teams to ship results. This is a research and modeling role, not product analytics.

The team is a small research lab in the Palo Alto office, a few minutes from Stanford, that looks a few years ahead and proves or drops risky ideas quickly.

Responsibilities

• Choose two or three research bets where the company can lead in AI on business data, agree on how success will be measured, and stop the bets that fail.
• Collaborate closely with top universities and institutes, including Stanford HAI.
• Follow emerging work such as relational and graph foundation models, neuro-symbolic agents and context graphs, test it on company data, and argue for the next bet.
• Develop methods that learn, align and evolve ontologies and knowledge graphs of business objects, rules and actions from metadata, master data, documents and process data, combining LLM-assisted ontology learning, schema matching, entity resolution and expert review.
• Build agents grounded in business semantics: graph-based retrieval, semantic parsing into SQL, SPARQL and OData calls (including finding the right API) across very large schemas, and checks that an agent's plan follows business rules and user permissions before it runs.
• Create graph learning and relational foundation models that use knowledge-graph structure to predict, explain and simulate business outcomes.
• Design benchmarks and simulation environments for agents on ERP data, since public benchmarks fall short, so the company can tell when AI is reliable enough to act.
• Work with data teams and domain experts in co-innovation projects, turning messy enterprise data problems into tractable ontology problems, and work with the teams behind the company's knowledge graph, business data platform and AI assistant to ship what works.
• Publish at top venues, release code and benchmarks where possible, co-lead university collaborations, and mentor scientists and PhD interns.
• In your first year, define two or three research directions agreed with leadership, at least one with a product partner and one with a longer horizon; get one result running in a product or pilot; submit papers and release a benchmark or open-source tool others use; and help other scientists do stronger work.

Requirements

• Required: 8+ years in applied research, data science or machine learning, or a PhD plus 5+ years, including research you led that changed a product or platform.
• Required: an MS or PhD in computer science, machine learning, statistics, mathematics or a related field, or equivalent practical experience.
• Required: deep expertise in one area and hands-on experience in another, with at least one in knowledge graphs, ontologies or semantic data integration.
• Required, knowledge graphs: RDF, OWL, SHACL, SPARQL or property graphs, ontology and schema matching, entity resolution, semantic layers, and you have made modeling calls yourself on entity boundaries, inheritance and relationship semantics, not only queried existing graphs.
• Required, machine learning: GNNs, relational deep learning, or relational and tabular foundation models.
• Required, LLM systems: retrieval, tool and API discovery, text-to-SQL and schema linking, tool-using agents, fine-tuning or post-training over structured data.
• Required: evaluation and benchmark design for AI systems.
• Required: strong machine learning and statistics foundations, fluent in Python and SQL, with PyTorch or JAX, and comfortable with schemas spanning thousands of related tables.
• Required: concrete examples of problem solving, such as turning an ambiguous business question into a testable research problem, starting with the simplest approach that could work, and stopping or redirecting work, including your own, when the evidence said so.
• Required: concrete examples of communication, such as explaining a trade-off to an executive concisely, a paper or memo that led a team outside yours to adopt, fund or drop something, and working through a design with skeptical engineers.
• Preferred: publications at venues such as NeurIPS, ICML, ICLR, KDD, ACL/EMNLP, SIGMOD/VLDB, ISWC or WWW, or widely used open-source software, datasets or semantic standards work.
• Preferred: LLM-assisted ontology learning, neuro-symbolic methods, agent memory, context graphs and decision traces, or reinforcement learning for tool-using agents.
• Preferred: process mining (including object-centric event data), causal inference, forecasting, or privacy-preserving and synthetic data.
• Preferred: familiarity with how enterprises run in finance, supply chain, procurement and HR, plus experience mentoring scientists or leading university collaborations.
• A plus, not a requirement: experience with the company's products.
• All roles require comfort using AI tools, including agentic AI, in daily work: writing effective inputs, automating routine tasks, judging the quality of AI output and driving adoption within a team.
• You do not need to match every line to apply.

Conditions

Hybrid: three days a week in the Palo Alto office.

Travel of up to 10%.

Principal-level individual contributor, comparable to Staff or Principal at other Bay Area technology companies. You will drive research, set standards, mentor scientists who do not report to you, represent the research externally and keep writing code.

The salary range is stated in the posting. Actual pay falls within it and depends on education, skills, experience, location and scope of role, as decided through selection.

Variable incentives may be offered under the relevant plan terms. Any incentive is target-based, with payouts depending on company and individual performance.

Use of AI in recruiting follows the company's ethical AI guidelines, and violations may lead to disqualification.

Applicant information is kept confidential under equal employment opportunity guidelines.

Benefits

• Constant learning and skill growth, with a team that wants you to grow and succeed.
• Great benefits.
• Permanent employees may use the employee referral program, which covers permanent roles only.