Agents for AI-Infra and Science (AIS) research
Optimizing machine learning infrastructure. Accelerating science.
About us
AIS is a group of scientists and engineers with the shared mission of accelerating scientific discovery and algorithmic optimization by seamlessly co-designing models and agent frameworks. We believe the next era of AI requires a unified approach: one that deeply optimizes the entire machine learning infrastructure stack by delivering breakthrough improvements in compute efficiency, memory management, and job scheduling, while simultaneously tackling complex problems in fundamental sciences like math and physics.
Machine Learning infra & code optimization
Designing autonomous multi-agent frameworks for code rewrite and optimization with applications including but not limited to kernel optimization, dynamic job scheduling, and large-scale, repo-level code optimization to maximize execution speed and efficiency across the entire hardware stack.
Verifiable reasoning & code generation
Building advanced agentic frameworks capable of rigorous, multi-step reasoning. By ensuring logical accuracy and formal verification, these systems unlock a new standard of highly reliable AI-generated code, and deliver rigorous proofs for complex math and physics.
Memory-efficient models and agents
Innovating model design and post-training techniques specifically tailored to reduce memory footprints while maintaining robust reasoning capabilities for long-horizon agentic tasks.
“Architecting the future of AI by innovating on model and agent design to solve the hardest problems in scientific discovery and system optimization.”
Featured research
Leap: Supercharging LLMs for Formal Mathematics with Agentic Frameworks
Po-Nien Kung, Linfeng Song, Dawsen Hwang, Jinsung Yoon, Chun-Liang Li, Simone Severini, Mirek Olšák, Edward Lockhart, Quoc V Le, Burak Gokturk, Thang Luong, Tomas Pfister, Nanyun Peng.
DeepCrossAttention: Supercharging Transformer Residual Connections
Mike Heddes, Adel Javanmard, Kyriakos Axiotis, Gang Fu, MohammadHossein Bateni, Vahab Mirrokni
-
The Network Structure of Mathlib
Xinze Li, Nanyun Peng, Simone Severini, Patrick Shafto
-
Nexus: An Agentic Framework for Time Series Forecasting
Sarkar Snigdha Sarathi Das, Palash Goyal, Mihir Parmar, Nanyun Peng, Vishy Tirumalashetty, Chun-Liang Li, Rui Zhang, Jinsung Yoon, Tomas Pfister
-
LEAF: A Living Benchmark for Event-Augmented Forecasting
Mingtian Tan, Mihir Parmar, Palash Goyal, Chun-Liang Li, Nanyun Peng, Thomas Hartvigsen, Jinsung Yoon, Tomas Pfister
-
PaperBanana-Interact: Scientific Diagram Refinement with Multi-Turn Human Feedback
Xueqing Wu, Ashwin Balasubramanian, Bingxuan Li, Dawei Zhu, Kai-Wei Chang, Yale Song, Yiwen Song, Rui Meng, Tomas Pfister, Nanyun Peng
Some of our people
Violet Peng
Lead, Senior Staff Research Scientist
Linfeng Song
Staff Research Engineer
Mike Heddes
Research Engineer
Locations
Bay Area
Greater Seattle Area