Enterprises store massive volumes of business-critical information in relational databases, but semantic access to this data remains limited by schema complexity and rigid query interfaces. A key research question we explore is whether we can construct relational embeddings across entities that capture the rich semantic meaning of business objects which are grounded not just in values, but in relationships, constraints, and structure inherent to relational data. Building on advances in relational learning, which has proven effective for predictive tasks such as classification and regression, we investigate how these embeddings can be extended to support semantic search, user intent understanding, and agentic AI workflows. Relational embeddings enable semantic indexing over structured data, unlocking faster and more precise search across large-scale, multimodal relational databases. As part of this effort, you will survey and prototype relational machine learning approaches, identify and curate suitable datasets for training and evaluating relational embedding techniques, and validate the most promising methods for scalability, robustness, and productization feasibility in enterprise-scale environments.
The person stepping into this role is expected to complete the internship using a wide and diverse set of skills.
- Good understanding of machine learning and AI
- Good understanding of LLMs, unsupervised learning and embedding models
- Experience with programming in Python
- Experience in working on research projects
- An understanding of internals of ML frameworks, LLMs, and training or fine-tuning ML models, is a plus
- Familiarity with SQL and an understanding of internals of database systems is a plus
Career Level - IC0