From-Scratch Paged KV-Cache with Block-Table Virtual Memory and Eviction Policies in Python
A hands‑on guide to building a paged KV‑cache in Python, complete with virtual memory, eviction policies, and benchmark results you can run today.
Implementing SPIFFE Workload Identity for Zero Trust Networks
SPIFFE replaces static credentials with dynamic mTLS for workloads. This article covers deployment patterns, service mesh integration, and hardening strategies for zero trust networks.
From-Scratch Mixture-of-Experts Routing Layer with Noisy Top‑k Gating and Load‑Balancing Auxiliary Loss
A complete guide and runnable PyTorch implementation of a Mixture‑of‑Experts layer with noisy top‑k routing, auxiliary load‑balancing, and token dispatch, perfect for a portfolio CV.
Hands‑On Build Guide: Minimal Mixture‑of‑Experts Inference Engine in Python
A step‑by‑step guide to coding a lightweight MoE inference engine in pure Python, complete with runnable code and extension ideas.
Optimizing ClickHouse MergeTree Compaction: Streaming Ingestion at High Throughput
Optimizing MergeTree compaction is critical for maintaining query performance under heavy streaming loads. This post covers part merging strategies, background pool tuning, and disk I/O management for production ClickHouse deployments.