CUDA / ML systems
Fused-Linear-Attention ↗
The repository behind the fused attention case study, focused on custom CUDA execution and profiling.
Code
Repositories that back up the case studies.
I do not treat GitHub as a dump of everything I have touched. The repositories here matter because they connect directly to the projects and notes elsewhere on the site.
Training, evaluation, deployment, and feedback-loop code for deadline and expiry extraction.
Recommendation
StyleSync ↗
FastAPI serving, retrieval workflows, and recommendation-system logic from the StyleSync project.