Scholar
Mansheej Paul
Google Scholar ID: gmZt3VgAAAAJ
Research Scientist, Databricks
Citations & Impact
Citations
1,563
H-index
11
i10-index
11
Publications
19
Co-authors
0
Publications
Resume
Academic Achievements
- Publications can be found on Google Scholar.
Research Experience
- Research scientist at Mosaic AI Research, Databricks; research intern at FAIR, Meta AI; worked on the science of deep learning through the lens of data, loss landscapes, and neural tangent kernels.
Education
- Ph.D. in Applied Physics from Stanford University, advised by Surya Ganguli.
Background
- Research interests span pre-training and post-training LLMs with a focus on optimizing data quality, distribution, and curricula. Currently building synthetic data pipelines to scale inference compute, create diverse generations, and develop strategies to verify and filter them into high-quality training data. Aims to create reliable, consistent, and trustworthy AI systems through rigorous evaluation of model behavior and how it is shaped by training data properties.
Miscellany
- Volunteered for SF New Deal, helping research and draft their economic impact report; enjoys social dancing, mostly West Coast Swing.