ENGINEERING RESEARCH NOTEBOOK
AI Inference & GPU Systems Engineer
Reproducible experiments in vLLM, CUDA, GPU serving, distributed inference, and Kubernetes—built from source code, measured artifacts, and explicit limitations.
FEATURED PROJECT
A compatibility and runtime-delivery toolkit for reproducible upstream vLLM inference on an explicitly validated Kaggle dual-NVIDIA-T4 environment.
Project scope and evidence →What a controlled dual-T4 study revealed about when TP=2 helps, when TP=1 wins, how workload shape changes the crossover, and why compatibility, capacity, performance, and quality must be measured separately.
An evidence-first introduction to the compatibility, provenance, and experiment design behind running upstream vLLM in a managed dual-T4 notebook environment.