THE BLOG · FREE
What changed in serving, and what it means in the interview.
Long-form articles on what changed in serving engines, kernels and hardware, and how interview rounds use it, each dated and with the arithmetic shown. Start with the question bank to practice on.
How to read an inference engine: a route through code, kernels and measurement
A practical reading route through Mini-SGLang, LLMs-from-scratch, Uzu, AIPerf and primary inference courses. Each stop has a specific file or section, an engineering question and an artifact to produce.
Vikas Mishra · Oct 21, 2026 · 10 min read
RMSNorm on an RTX 5070: why more reduction threads did not win every shape
An original FP32 CUDA investigation with 48 correctness cases, two repeated timing runs, raw samples and runnable kernels. Compare 32 and 256 threads per row without treating a desktop microbenchmark as serving performance.
Vikas Mishra · Oct 21, 2026 · 9 min read
Done reading? Start practicing.
The must-know inference questions →