Reasoning and distillation
Training methods that help language models identify and improve flawed reasoning.
Applied Scientist
I lead research and engineering for efficient, reliable language models.
I work across language model reasoning, efficient model systems, and trustworthy evaluation, connecting research ideas with practical engineering.
News
Selected updates
Research themes
Questions that shape my work
Efficient model systems
Model design and adaptation methods that make capable systems more efficient to use.
Trustworthy evaluation
Evaluation and safeguards that make model behavior easier to assess and rely on.
Selected publications
Research in practice
2026 · Featured research
Critique-Guided Distillation for Robust Reasoning via Refinement
International Conference on Machine Learning (ICML 2026) · Proceedings · Published
Critique-Guided Distillation trains a student to refine flawed responses using teacher critiques as training-only supervision, without requiring critiques at inference.
Scroll the figure sideways to read its labels, or open the full-size image below.
Lead figure
Teach the student to act on feedback during training so it can answer in one pass at inference.
A student learns from teacher critiques and refined answers during training, then answers without a teacher at inference.
Source: Berkcan Kapusuzoglu et al., Critique-Guided Distillation (2026), Figure 1. License: CC BY 4.0; rasterized without scientific edits.
Open full-size figure for labels and details.
2026 · Featured research
When Load-Balancing Goes Too Far: Expert Pruning in Over-Dispersed Mixture-of-Experts Models
NeurIPS 2026 Workshop on On-Device Intelligence · Workshop · Accepted
When load balancing spreads routing too uniformly across experts, router scores become less useful for deciding which experts to prune. Domain-aware score allocation accounts for which capabilities each expert supports.
2025 · Featured research
SPEAR-MM: Selective Parameter Evaluation and Restoration via Model Merging for Efficient Financial LLM Adaptation
2025 IEEE International Conference on Big Data · Conference · Published
SPEAR-MM uses post-hoc layer evaluation and spherical interpolation merging to preserve general capabilities during financial-domain adaptation of language models.
Career arc
Research grounded in engineering
My work has progressed from scientific, physics-informed machine learning to language-model research, and now to research-led production AI leadership: carrying rigorous ideas through evaluation and into reliable systems.
Education
- Vanderbilt University - Ph.D. in Civil Engineering
- University of Erlangen-Nuremberg — M.S. in Computational Engineering
- Delft University of Technology — M.S. in Applied Mathematics
- Bilkent University — B.S. in Mechanical Engineering
Contact