DBIS

Yixin Peng (he/him)

I research AI systems that learn from connected data, reason with evidence, and improve through feedback.

As a doctoral researcher at RWTH Aachen University, I work at the intersection of relational deep learning, temporal heterogeneous graphs, and graph–language models. My goal is to turn the rich information in interconnected databases into useful predictions and understandable, evidence-backed answers.

I am working toward foundation models for relational data that can handle new tasks without task-specific fine-tuning. This research brings together graph representation learning, multimodal pretraining, and language model post-training to support evidence-grounded prediction and reasoning.

I also explore recursively self-improving multi-agent systems, investigating how iterative feedback and formal reasoning can help AI systems refine their problem-solving processes while keeping their outputs traceable and verifiable.

Across these directions, I aim to build AI that learns from structured relationships, supports its answers with evidence, and improves in ways we can evaluate.