Rik Koncel-Kedziorski
ML Scientist, Apple
My research builds AI systems that reason about the people they interact with. I work on methods for interpersonal reasoning, including training language models to understand the psychological states, beliefs, and preferences underlying observed user behavior with the broader goal of improving human-AI collaboration.
This builds on a decade of work on reasoning in AI. Early in my career I focused on formal reasoning, developing foundational datasets and methods for mathematical problem solving (MAWPS, MathQA) and scientific text generation (Text Generation from Knowledge Graphs with Graph Transformers, Explaining Relationships Between Scientific Documents). In industry I extended these ideas to financial reasoning (DocFinQA, BizBench) and to one of the first production-scale generative QA systems at Alexa. My current work brings these threads together by grounding LLM reasoning in validated psychological theory — see PrimeX and Improving Language Model Personas via Rationalization with Psychological Scaffolds (both EMNLP 2025).
I completed my PhD in Computational Linguistics at the University of Washington in 2019, advised by Hanna Hajishirzi and Gina-Anne Levow, and continued as a postdoctoral researcher with Noah Smith at UW CSE. Before Apple, I was a Research Scientist at S&P Global Kensho and an Applied Scientist at Amazon.
news
| Feb 12, 2026 | Invited talk at the Penn Primals Lab on psychological theories of personality in AI. |
|---|---|
| Nov 06, 2025 | Gave a keynote at the PALS (Personalization of Language Systems) workshop at EMNLP 2025. |
| Nov 05, 2025 | Improving Language Model Personas via Rationalization with Psychological Scaffolds accepted at EMNLP 2025 (with Brihi Joshi, Xiang Ren, Swabha Swayamdipta, Tim Paek). |
| Nov 05, 2025 | PrimeX: A Dataset of Worldview, Opinion, and Explanation accepted at EMNLP 2025. |
| Aug 01, 2024 | DocFinQA: A Long-Context Financial Reasoning Dataset accepted at ACL 2024. |