Agent-Centric Data Systems

Internship or Master's Thesis
Agent-Centric Data Systems
Ref. 2026_014

Context

Enterprise data systems are evolving from pipeline-centric architectures towards agent-centric systems, in which large language model (LLM)-based agents interpret user intent and execute data operations on behalf of the user. In contrast to traditional approaches based on predefined ETL pipelines and static orchestration, these systems rely on adaptive, multi-step reasoning and dynamic tool usage.

This shift introduces a number of fundamental challenges. As the number of available tools and task-specific capabilities increases, efficiently managing the agent’s context becomes critical. Furthermore, established evaluation methodologies based on single-query accuracy are no longer sufficient to assess the quality of multi-step, agent-driven workflows. In addition, enabling agents to leverage prior interactions in order to improve performance and tailor behavior to individual users remains an open research problem.

Our group develops next-generation data systems that address these challenges at the intersection of LLMs, data management, and systems engineering. Our current focus lies on data-centric operations, including synthetic data generation, autonomous data analysis, and cross-engine data retrieval. At the same time, the underlying system design aims to generalize beyond data management and support a broader class of agent-driven applications.

Your Role

You will contribute to the design, implementation, and evaluation of agent-centric data systems. The position offers flexibility in shaping the specific focus area, while maintaining a strong connection to ongoing system development and research activities. Typical areas of work include:

Designing and implementing scalable approaches for matching user intent to relevant tools and capabilities in environments with a large and continuously evolving set of available functionalities
Developing mechanisms for cross-session learning, enabling the agent to capture, structure, and reuse operational knowledge (e.g., provider-specific behavior, effective patterns, and workarounds) in order to improve efficiency and adapt to individual users
Investigating agent-driven execution across heterogeneous data systems, where a single user request requires coordinated operations spanning multiple types of data management platforms
Designing and evaluating extensible plugin interfaces that allow third-party data capabilities to be discovered and integrated dynamically at runtime
Establishing evaluation frameworks for agent-based systems, including the generation of synthetic data and the definition of metrics that capture the correctness and robustness of multi-step workflows beyond isolated query accuracy

The position provides the opportunity to work on both system-oriented contributions and empirical evaluation. Depending on the scope of the project, results may contribute to product development as well as to publications at leading conferences.

Minimum Qualifications

Master’s student in Computer Science, Data Science, or a related discipline
Strong software engineering skills with solid experience in Python
Good understanding of relational databases and SQL
Familiarity with large language models and their practical application
Experience with version control systems (e.g., Git) and collaborative development workflows
Ability to work independently on open-ended problems
Strong written and verbal communication skills in English

Preferred Qualifications

Experience with LLM-based systems or agent frameworks (e.g., Model Context Protocol or similar concepts)
Familiarity with modern data systems and tools (e.g., DuckDB, Apache Arrow, Spark)
Background in data engineering, query processing, or data integration
Experience with benchmarking and evaluation of complex systems
Interest in combining system development with research-oriented work

What We Offer

The opportunity to work on cutting-edge systems at the intersection of LLMs and data management
Close collaboration with experienced researchers and engineers in a research-driven environment
Access to modern LLM infrastructure and heterogeneous data platforms
portunities to contribute to open-source projects and research publications
A flexible and collaborative working environment within an international team

Diversity & Work Environment
IBM is committed to fostering diversity and inclusion in the workplace. You will join an open, multicultural research environment that values different perspectives and supports flexible working arrangements. Our goal is to help all genders and backgrounds thrive professionally while maintaining a healthy work–life balance.

How to Apply

Please submit your CV, a brief statement of interest (maximum one page), and your academic transcript. If available, include links to relevant projects or code repositories.

Apply