KernelBlaster: Continual Cross-Task CUDA Optimization via Memory-Augmented In-Context Reinforcement Learning
Under Submission
Kris Shengjun Dong, Sahil Modi, Dima Nikiforov, Sana Damani, Edward Lin, Siva Kumar Sastry Hari, Christos Kozyrakis.
Hi there, I am Kris!
My research focuses on systems for AI and AI for systems, with an emphasis on LLM-driven optimization for efficient, high-performance execution from the kernel level to the full system stack. I develop LLM-based agentic flows for automating code generation and performance–power optimization, and integrate these methods into compilers and runtime systems for heterogeneous and edge architectures.
More broadly, I build infrastructure frameworks that support end-to-end hardware–software co-design, enabling rapid experimentation and cross-stack integration, while providing full-stack performance characterization and closed-loop performance optimization and validation. I am passionate about open-source ecosystems and reproducible research. Two of my main works have been recognized with the Distinguished Artifact Award at ISCA 2023 and the Distinguished Artifact Award at IISWC 2025.
