


Latest move
Anthropic → Mirendil · Founding member
Path
Current
MirendilNEWFounding member
Previous
Anthropic– 2026Co-Leading AI Scientist, Discovery Team- Meta2025Student Researcher
Google2022 – 2025Student Researcher
Microsoft Research2019Undergraduate researcher
University of Texas at AustinPhD (fifth year), advised by Inderjit Dhillon- BIBITS Pilani, GoaBachelor's, advised by Ravishankar Krishnaswamy (undergraduate thesis)
Papers
- The Art of Scaling Reinforcement Learning Compute for LLMsarXiv 2025Studies how reinforcement learning performance of language models changes as training compute grows, and proposes a recipe (ScaleRL).
- MatFormer: Nested Transformer for Elastic InferencearXiv 2023Trains one Transformer containing nested smaller models that can be extracted for different compute budgets.
- Dual-Encoders for Extreme Multi-Label ClassificationarXiv 2023Studies how dual-encoder models can be trained for classification over very large label sets.