Post-training data
Scalable data methods for improving instruction following, grounded behavior, reasoning, and product quality.
dataevaluation
Welcome!
I am a machine learning researcher working on large language models, with a focus on improving their reliability, controllability, and alignment with human intent.
I am currently a Principal Applied Scientist at Microsoft, where I work on Responsible AI and alignment. My work develops techniques for improving the behavior, reasoning, and safety of large-scale models.
From 2022 to early 2026, I was a Senior Applied Scientist at Amazon, most recently in Amazon AGI. I worked on large-scale post-training, with an emphasis on reasoning, controllability, and product quality across deployed applications. That work spanned deep-research and agentic search systems, retrieval-augmented generation, controllable generation, and faithful summarization, including GaRAGe, FactGraph, and controllable-readability summarization.
I completed my Ph.D. in computer science at the Technical University of Darmstadt in the UKP Lab advised by Iryna Gurevych, in 2022. My thesis studied structured representations and graph-based methods for natural language generation. Earlier, I worked on graph representation learning with Daniel R. Figueiredo at UFRJ, including struc2vec.
Research focus
Alignment and safety
Methods and evaluations for selective refusal, safe responses under uncertainty, and aligned model behavior.
refusalsafety
Grounded reasoning
RAG and agentic search systems that connect generation to evidence, source attribution, and faithful synthesis.
RAGagents
Controllable generation
Techniques for shaping style, readability, summaries, and task behavior while preserving factual consistency.
controlquality
Professional service
- Senior Area Chair, NAACL 2025
- Tutorial Chair, ACL 2024
- Area Chair and reviewer across ACL, EACL, EMNLP, COLING, COLM, ARR, AAAI, and related workshops
- Reviewer for journals including JMLR and Language Resources and Evaluation